auto_round format

Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>

config.json

@@ -45,12 +45,11 @@
   "quantization_config": {
     "amp": true,
     "autoround_version": "0.4.2.dev",
+    "backend": "auto_round:gptq:exllamav2",
     "batch_size": 8,
     "bits": 4,
-    "block_name_to_quantize": "transformer.encoder.layers",
-    "damp_percent": 0.01,
     "data_type": "int",
-    "desc_act": false,
+    "dataset": "NeelNanda/pile-10k",
     "enable_minmax_tuning": true,
     "enable_norm_bias_tuning": false,
     "enable_quanted_input": true,
@@ -61,12 +60,11 @@
     "lr": 0.001,
     "minmax_lr": 0.001,
     "nsamples": 512,
-    "quant_method": "gptq",
+    "quant_method": "intel/auto-round",
     "scale_dtype": "torch.float16",
     "seqlen": 2048,
     "sym": true,
-    "to_quant_block_names": "transformer.encoder.layers",
-    "true_sequential": false
+    "to_quant_block_names": "transformer.encoder.layers"
   },
   "rmsnorm": true,
   "rope_ratio": 1,

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62829b26f86cde3129ce4012e21dd67e0259bc05131c3a463b1dfb854f680949
-size 4975739040
+oid sha256:d7c295fef9b3f7fd900bdf38512c4a1a05621e85a3123de63d2dbb2ad4139b40
+size 4973226600

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94f8b1740de71c38f3d7968e6c29191dbe7231dcafb898d6fbc16348efe2a203
-size 4984273648
+oid sha256:50d9300df8c11f3332a3bec54fbe5f3da7d2a51cd76152a5d4a89229d6b20fcc
+size 4983924232

model.safetensors.index.json

@@ -1,23 +1,20 @@
 {
   "metadata": {
-    "total_size": 15742098496
+    "total_size": 15739251776
   },
   "weight_map": {
     "transformer.embedding.word_embeddings.weight": "model-00001-of-00004.safetensors",
     "transformer.encoder.final_layernorm.weight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.0.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.0.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.0.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.0.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -28,18 +25,15 @@
     "transformer.encoder.layers.0.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.0.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.1.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.1.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.1.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -50,18 +44,15 @@
     "transformer.encoder.layers.1.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.1.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.10.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.10.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.10.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -72,18 +63,15 @@
     "transformer.encoder.layers.10.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.10.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.11.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.11.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.11.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -94,18 +82,15 @@
     "transformer.encoder.layers.11.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.11.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.12.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.12.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.12.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -116,18 +101,15 @@
     "transformer.encoder.layers.12.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.12.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.13.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.13.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.13.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -138,18 +120,15 @@
     "transformer.encoder.layers.13.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.13.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.14.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.14.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.14.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -160,18 +139,15 @@
     "transformer.encoder.layers.14.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.14.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.15.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.15.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.15.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -182,18 +158,15 @@
     "transformer.encoder.layers.15.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.15.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.16.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.16.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.16.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -204,18 +177,15 @@
     "transformer.encoder.layers.16.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.16.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.17.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.17.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.17.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -226,18 +196,15 @@
     "transformer.encoder.layers.17.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.17.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.18.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.18.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.18.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -248,18 +215,15 @@
     "transformer.encoder.layers.18.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.18.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.19.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.19.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.19.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -270,18 +234,15 @@
     "transformer.encoder.layers.19.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.19.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.2.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.2.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.2.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -292,18 +253,15 @@
     "transformer.encoder.layers.2.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.2.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.20.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.20.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.20.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -314,18 +272,15 @@
     "transformer.encoder.layers.20.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.20.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.21.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.21.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.21.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -336,18 +291,15 @@
     "transformer.encoder.layers.21.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.21.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.22.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.22.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.22.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -358,18 +310,15 @@
     "transformer.encoder.layers.22.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.22.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.23.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.23.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.23.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -380,18 +329,15 @@
     "transformer.encoder.layers.23.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.23.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.24.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.24.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.24.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -402,18 +348,15 @@
     "transformer.encoder.layers.24.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.24.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.25.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.25.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.25.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -424,18 +367,15 @@
     "transformer.encoder.layers.25.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.25.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.26.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.26.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.26.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -446,18 +386,15 @@
     "transformer.encoder.layers.26.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.26.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.27.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.27.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.27.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -468,18 +405,15 @@
     "transformer.encoder.layers.27.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.27.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.28.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.28.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.28.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -490,18 +424,15 @@
     "transformer.encoder.layers.28.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.28.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.29.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.29.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.29.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -512,18 +443,15 @@
     "transformer.encoder.layers.29.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.29.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.3.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.3.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.3.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -534,18 +462,15 @@
     "transformer.encoder.layers.3.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.3.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.30.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.30.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.30.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -556,18 +481,15 @@
     "transformer.encoder.layers.30.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.30.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.31.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.31.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.31.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -578,18 +500,15 @@
     "transformer.encoder.layers.31.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.31.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.32.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.32.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.32.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -600,18 +519,15 @@
     "transformer.encoder.layers.32.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.32.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.33.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.33.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.33.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -622,18 +538,15 @@
     "transformer.encoder.layers.33.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.33.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.34.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.34.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.34.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -644,18 +557,15 @@
     "transformer.encoder.layers.34.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.34.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.35.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.35.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_4h_to_h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_4h_to_h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_4h_to_h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_4h_to_h.scales": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.35.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_h_to_4h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_h_to_4h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_h_to_4h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.mlp.dense_h_to_4h.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.35.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.35.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.35.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.35.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.35.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -666,18 +576,15 @@
     "transformer.encoder.layers.35.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.35.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.36.input_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.36.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_4h_to_h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_4h_to_h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_4h_to_h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_4h_to_h.scales": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.36.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_h_to_4h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_h_to_4h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_h_to_4h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.mlp.dense_h_to_4h.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.36.self_attention.dense.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.self_attention.dense.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.self_attention.dense.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.self_attention.dense.qzeros": "model-00002-of-00004.safetensors",
@@ -688,18 +595,15 @@
     "transformer.encoder.layers.36.self_attention.query_key_value.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.36.self_attention.query_key_value.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.input_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.37.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_4h_to_h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_4h_to_h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_4h_to_h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_4h_to_h.scales": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.37.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_h_to_4h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_h_to_4h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_h_to_4h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.mlp.dense_h_to_4h.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.37.self_attention.dense.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.self_attention.dense.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.self_attention.dense.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.self_attention.dense.qzeros": "model-00002-of-00004.safetensors",
@@ -710,18 +614,15 @@
     "transformer.encoder.layers.37.self_attention.query_key_value.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.37.self_attention.query_key_value.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.input_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.38.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_4h_to_h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_4h_to_h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_4h_to_h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_4h_to_h.scales": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.38.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_h_to_4h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_h_to_4h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_h_to_4h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.mlp.dense_h_to_4h.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.38.self_attention.dense.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.self_attention.dense.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.self_attention.dense.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.self_attention.dense.qzeros": "model-00002-of-00004.safetensors",
@@ -732,18 +633,15 @@
     "transformer.encoder.layers.38.self_attention.query_key_value.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.38.self_attention.query_key_value.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.input_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.39.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_4h_to_h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_4h_to_h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_4h_to_h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_4h_to_h.scales": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.39.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_h_to_4h.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_h_to_4h.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_h_to_4h.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.mlp.dense_h_to_4h.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
-    "transformer.encoder.layers.39.self_attention.dense.bias": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.self_attention.dense.g_idx": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.self_attention.dense.qweight": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.self_attention.dense.qzeros": "model-00002-of-00004.safetensors",
@@ -754,18 +652,15 @@
     "transformer.encoder.layers.39.self_attention.query_key_value.qzeros": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.39.self_attention.query_key_value.scales": "model-00002-of-00004.safetensors",
     "transformer.encoder.layers.4.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.4.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.4.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.4.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -776,18 +671,15 @@
     "transformer.encoder.layers.4.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.4.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.5.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.5.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.5.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -798,18 +690,15 @@
     "transformer.encoder.layers.5.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.5.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.6.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.6.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.6.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -820,18 +709,15 @@
     "transformer.encoder.layers.6.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.6.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.7.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.7.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.7.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -842,18 +728,15 @@
     "transformer.encoder.layers.7.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.7.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.8.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.8.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.8.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",
@@ -864,18 +747,15 @@
     "transformer.encoder.layers.8.self_attention.query_key_value.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.8.self_attention.query_key_value.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.input_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.9.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_4h_to_h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_4h_to_h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_4h_to_h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_4h_to_h.scales": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.9.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_h_to_4h.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_h_to_4h.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_h_to_4h.qzeros": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.mlp.dense_h_to_4h.scales": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
-    "transformer.encoder.layers.9.self_attention.dense.bias": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.self_attention.dense.g_idx": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.self_attention.dense.qweight": "model-00001-of-00004.safetensors",
     "transformer.encoder.layers.9.self_attention.dense.qzeros": "model-00001-of-00004.safetensors",

modeling_chatglm.py

@@ -0,0 +1,1329 @@
+""" PyTorch GLM-4V model. """
+import math
+import sys
+import torch
+import torch.utils.checkpoint
+import torch.nn.functional as F
+from torch import nn
+from torch.nn import CrossEntropyLoss, LayerNorm, MSELoss, BCEWithLogitsLoss
+from torch.nn.utils import skip_init
+from typing import Optional, Tuple, Union, List, Dict, Any
+
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+    SequenceClassifierOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging, is_torch_npu_available
+from transformers.generation.logits_process import LogitsProcessor
+from transformers.generation.utils import LogitsProcessorList, StoppingCriteriaList, GenerationConfig, ModelOutput
+
+from .visual import EVA2CLIPModel
+from .configuration_chatglm import ChatGLMConfig
+
+try:
+    from transformers.utils import is_flash_attn_greater_or_equal_2_10, is_flash_attn_2_available
+
+    if is_flash_attn_2_available():
+        from flash_attn import flash_attn_func, flash_attn_varlen_func
+        from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
+except:
+    pass
+
+# flags required to enable jit fusion kernels
+
+if sys.platform != 'darwin' and not is_torch_npu_available():
+    torch._C._jit_set_profiling_mode(False)
+    torch._C._jit_set_profiling_executor(False)
+    torch._C._jit_override_can_fuse_on_cpu(True)
+    torch._C._jit_override_can_fuse_on_gpu(True)
+
+logger = logging.get_logger(__name__)
+
+LANGUAGE_TOKEN_TYPE = 0
+VISION_TOKEN_TYPE = 1
+
+_CHECKPOINT_FOR_DOC = "THUDM/ChatGLM"
+_CONFIG_FOR_DOC = "ChatGLMConfig"
+
+
+def default_init(cls, *args, **kwargs):
+    return cls(*args, **kwargs)
+
+
+class InvalidScoreLogitsProcessor(LogitsProcessor):
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
+        if torch.isnan(scores).any() or torch.isinf(scores).any():
+            scores.zero_()
+            scores[..., 198] = 5e4
+        return scores
+
+
+class PrefixEncoder(torch.nn.Module):
+    """
+    The torch.nn model to encode the prefix
+    Input shape: (batch-size, prefix-length)
+    Output shape: (batch-size, prefix-length, 2*layers*hidden)
+    """
+
+    def __init__(self, config: ChatGLMConfig):
+        super().__init__()
+        self.prefix_projection = config.prefix_projection
+        if self.prefix_projection:
+            # Use a two-layer MLP to encode the prefix
+            kv_size = config.num_layers * config.kv_channels * config.multi_query_group_num * 2
+            self.embedding = torch.nn.Embedding(config.pre_seq_len, kv_size)
+            self.trans = torch.nn.Sequential(
+                torch.nn.Linear(kv_size, config.hidden_size),
+                torch.nn.Tanh(),
+                torch.nn.Linear(config.hidden_size, kv_size)
+            )
+        else:
+            self.embedding = torch.nn.Embedding(config.pre_seq_len,
+                                                config.num_layers * config.kv_channels * config.multi_query_group_num * 2)
+
+    def forward(self, prefix: torch.Tensor):
+        if self.prefix_projection:
+            prefix_tokens = self.embedding(prefix)
+            past_key_values = self.trans(prefix_tokens)
+        else:
+            past_key_values = self.embedding(prefix)
+        return past_key_values
+
+
+def split_tensor_along_last_dim(
+        tensor: torch.Tensor,
+        num_partitions: int,
+        contiguous_split_chunks: bool = False,
+) -> List[torch.Tensor]:
+    """Split a tensor along its last dimension.
+
+    Arguments:
+        tensor: input tensor.
+        num_partitions: number of partitions to split the tensor
+        contiguous_split_chunks: If True, make each chunk contiguous
+                                 in memory.
+
+    Returns:
+        A list of Tensors
+    """
+    # Get the size and dimension.
+    last_dim = tensor.dim() - 1
+    last_dim_size = tensor.size()[last_dim] // num_partitions
+    # Split.
+    tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
+    # Note: torch.split does not create contiguous tensors by default.
+    if contiguous_split_chunks:
+        return tuple(chunk.contiguous() for chunk in tensor_list)
+
+    return tensor_list
+
+
+class RotaryEmbedding(nn.Module):
+    def __init__(self, dim, rope_ratio=1, original_impl=False, device=None, dtype=None):
+        super().__init__()
+        inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2, device=device).to(dtype=dtype) / dim))
+        self.register_buffer("inv_freq", inv_freq)
+        self.dim = dim
+        self.original_impl = original_impl
+        self.rope_ratio = rope_ratio
+
+    def impl(self, seq_length: int, dim: int, device: torch.device, dtype: torch.dtype):
+        base = 10000 * self.rope_ratio
+        inv_freq = 1.0 / (
+                base ** (torch.arange(0, dim, 2, device=device, dtype=torch.float32) / dim))
+        seq = torch.arange(seq_length, device=inv_freq.device, dtype=torch.float32)
+        freqs = torch.outer(seq, inv_freq)
+        # first part even vector components, second part odd vector components,
+        #  2 * dim in dimension size
+        emb = torch.cat((freqs, freqs), dim=-1)
+        return emb
+
+    def forward_impl(
+            self, seq_len: int, n_elem: int, dtype: torch.dtype, device: torch.device, base: int = 10000
+    ):
+        """Enhanced Transformer with Rotary Position Embedding.
+
+        Derived from: https://github.com/labmlai/annotated_deep_learning_paper_implementations/blob/master/labml_nn/
+        transformers/rope/__init__.py. MIT License:
+        https://github.com/labmlai/annotated_deep_learning_paper_implementations/blob/master/license.
+        """
+        # $\Theta = {\theta_i = 10000^{\frac{2(i-1)}{d}}, i \in [1, 2, ..., \frac{d}{2}]}$
+        base = base * self.rope_ratio
+        theta = 1.0 / (base ** (torch.arange(0, n_elem, 2, dtype=torch.float, device=device) / n_elem))
+
+        # Create position indexes `[0, 1, ..., seq_len - 1]`
+        seq_idx = torch.arange(seq_len, dtype=torch.float, device=device)
+
+        # Calculate the product of position index and $\theta_i$
+        idx_theta = torch.outer(seq_idx, theta).float()
+
+        cache = torch.stack([torch.cos(idx_theta), torch.sin(idx_theta)], dim=-1)
+
+        # this is to mimic the behaviour of complex32, else we will get different results
+        if dtype in (torch.float16, torch.bfloat16, torch.int8):
+            cache = cache.bfloat16() if dtype == torch.bfloat16 else cache.half()
+        return cache
+
+    def forward(self, max_seq_len, offset=0):
+        if self.original_impl:
+            return self.forward_impl(
+                max_seq_len, self.dim, dtype=self.inv_freq.dtype, device=self.inv_freq.device
+            )
+        else:
+            return self.impl(max_seq_len, self.dim, dtype=self.inv_freq.dtype, device=self.inv_freq.device)
+
+
+@torch.jit.script
+def apply_rotary_pos_emb(x: torch.Tensor, rope_cache: torch.Tensor) -> torch.Tensor:
+    # x: [b, np, sq, hn]
+    b, np, sq, hn = x.size(0), x.size(1), x.size(2), x.size(3)
+    rot_dim = rope_cache.shape[-2] * 2
+    x, x_pass = x[..., :rot_dim], x[..., rot_dim:]
+    # truncate to support variable sizes
+    rope_cache = rope_cache[:, :sq]
+    xshaped = x.reshape(b, np, sq, rot_dim // 2, 2)
+    rope_cache = rope_cache.view(-1, 1, sq, xshaped.size(3), 2)
+    x_out2 = torch.stack(
+        [
+            xshaped[..., 0] * rope_cache[..., 0] - xshaped[..., 1] * rope_cache[..., 1],
+            xshaped[..., 1] * rope_cache[..., 0] + xshaped[..., 0] * rope_cache[..., 1],
+        ],
+        -1,
+    )
+    x_out2 = x_out2.flatten(3)
+    return torch.cat((x_out2, x_pass), dim=-1)
+
+
+class RMSNorm(torch.nn.Module):
+    def __init__(self, normalized_shape, eps=1e-5, device=None, dtype=None, **kwargs):
+        super().__init__()
+        self.weight = torch.nn.Parameter(torch.empty(normalized_shape, device=device, dtype=dtype))
+        self.eps = eps
+
+    def forward(self, hidden_states: torch.Tensor):
+        input_dtype = hidden_states.dtype
+        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.eps)
+
+        return (self.weight * hidden_states).to(input_dtype)
+
+
+
+class CoreAttention(torch.nn.Module):
+    def __init__(self, config: ChatGLMConfig, layer_number):
+        super(CoreAttention, self).__init__()
+
+        self.apply_query_key_layer_scaling = config.apply_query_key_layer_scaling
+        self.attention_softmax_in_fp32 = config.attention_softmax_in_fp32
+        if self.apply_query_key_layer_scaling:
+            self.attention_softmax_in_fp32 = True
+        self.layer_number = max(1, layer_number)
+
+        projection_size = config.kv_channels * config.num_attention_heads
+
+        # Per attention head and per partition values.
+        self.hidden_size_per_partition = projection_size
+        self.hidden_size_per_attention_head = projection_size // config.num_attention_heads
+        self.num_attention_heads_per_partition = config.num_attention_heads
+
+        coeff = None
+        self.norm_factor = math.sqrt(self.hidden_size_per_attention_head)
+        if self.apply_query_key_layer_scaling:
+            coeff = self.layer_number
+            self.norm_factor *= coeff
+        self.coeff = coeff
+
+        self.attention_dropout = torch.nn.Dropout(config.attention_dropout)
+
+    def forward(self, query_layer, key_layer, value_layer, attention_mask):
+        pytorch_major_version = int(torch.__version__.split('.')[0])
+        if pytorch_major_version >= 2:
+            if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
+                context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
+                                                                                 is_causal=True)
+            else:
+                if attention_mask is not None:
+                    attention_mask = ~attention_mask
+                context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
+                                                                                 attention_mask)
+            context_layer = context_layer.transpose(1, 2).contiguous()
+            new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
+            context_layer = context_layer.reshape(*new_context_layer_shape)
+        else:
+            # Raw attention scores
+
+            # [b, np, sq, sk]
+            output_size = (query_layer.size(0), query_layer.size(1), query_layer.size(2), key_layer.size(2))
+
+            # [b, np, sq, hn] -> [b * np, sq, hn]
+            query_layer = query_layer.view(output_size[0] * output_size[1], output_size[2], -1)
+            # [b, np, sk, hn] -> [b * np, sk, hn]
+            key_layer = key_layer.view(output_size[0] * output_size[1], output_size[3], -1)
+
+            # preallocting input tensor: [b * np, sq, sk]
+            matmul_input_buffer = torch.empty(
+                output_size[0] * output_size[1], output_size[2], output_size[3], dtype=query_layer.dtype,
+                device=query_layer.device
+            )
+
+            # Raw attention scores. [b * np, sq, sk]
+            matmul_result = torch.baddbmm(
+                matmul_input_buffer,
+                query_layer,  # [b * np, sq, hn]
+                key_layer.transpose(1, 2),  # [b * np, hn, sk]
+                beta=0.0,
+                alpha=(1.0 / self.norm_factor),
+            )
+
+            # change view to [b, np, sq, sk]
+            attention_scores = matmul_result.view(*output_size)
+
+            # ===========================
+            # Attention probs and dropout
+            # ===========================
+
+            # attention scores and attention mask [b, np, sq, sk]
+            if self.attention_softmax_in_fp32:
+                attention_scores = attention_scores.float()
+            if self.coeff is not None:
+                attention_scores = attention_scores * self.coeff
+            if attention_mask is None and attention_scores.shape[2] == attention_scores.shape[3]:
+                attention_mask = torch.ones(output_size[0], 1, output_size[2], output_size[3],
+                                            device=attention_scores.device, dtype=torch.bool)
+                attention_mask.tril_()
+                attention_mask = ~attention_mask
+            if attention_mask is not None:
+                attention_scores = attention_scores.masked_fill(attention_mask, float("-inf"))
+            attention_probs = F.softmax(attention_scores, dim=-1)
+            attention_probs = attention_probs.type_as(value_layer)
+
+            # This is actually dropping out entire tokens to attend to, which might
+            # seem a bit unusual, but is taken from the original Transformer paper.
+            attention_probs = self.attention_dropout(attention_probs)
+            # =========================
+            # Context layer. [sq, b, hp]
+            # =========================
+
+            # value_layer -> context layer.
+            # [sk, b, np, hn] --> [b, np, sq, hn]
+
+            # context layer shape: [b, np, sq, hn]
+            output_size = (value_layer.size(1), value_layer.size(2), query_layer.size(0), value_layer.size(3))
+            # change view [b * np, sk, hn]
+            value_layer = value_layer.view(output_size[0] * output_size[1], value_layer.size(2), -1)
+            # change view [b * np, sq, sk]
+            attention_probs = attention_probs.view(output_size[0] * output_size[1], output_size[2], -1)
+            # matmul: [b * np, sq, hn]
+            context_layer = torch.bmm(attention_probs, value_layer)
+            # change view [b, np, sq, hn]
+            context_layer = context_layer.view(*output_size)
+            # [b, np, sq, hn] --> [b, sq, np, hn]
+            context_layer = context_layer.transpose(1, 2).contiguous()
+            # [b, sq, np, hn] --> [b, sq, hp]
+            new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
+            context_layer = context_layer.reshape(*new_context_layer_shape)
+
+        return context_layer
+
+class SdpaAttention(CoreAttention):
+    def forward(self, query_layer, key_layer, value_layer, attention_mask):
+        if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
+            context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
+                                                                             is_causal=True,
+                                                                             dropout_p=self.config.attention_dropout if self.training else 0.0)
+        else:
+            if attention_mask is not None:
+                attention_mask = ~attention_mask
+            context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
+                                                                             attention_mask,
+                                                                             dropout_p=self.config.attention_dropout if self.training else 0.0)
+        context_layer = context_layer.transpose(1, 2).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
+        context_layer = context_layer.reshape(*new_context_layer_shape)
+        return context_layer
+
+
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
+# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2
+class FlashAttention2(CoreAttention):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(self, query_states, key_states, value_states, attention_mask):
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+        batch_size, query_length = query_states.shape[:2]
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+        dropout = self.config.attention_dropout if self.training else 0.0
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            attn_output_unpad = flash_attn_varlen_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_in_batch_q,
+                max_seqlen_k=max_seqlen_in_batch_k,
+                dropout_p=dropout,
+                softmax_scale=None,
+                causal=causal,
+            )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            attn_output = flash_attn_func(
+                query_states, key_states, value_states, dropout, softmax_scale=None, causal=causal
+            )
+        attn_output = attn_output.reshape(batch_size, query_length, self.hidden_size_per_partition).contiguous()
+        return attn_output
+
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
+
+        key_layer = index_first_axis(
+            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        value_layer = index_first_axis(
+            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, self.num_attention_heads_per_partition, head_dim),
+                indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+CORE_ATTENTION_CLASSES = {
+    "eager": CoreAttention,
+    "sdpa": SdpaAttention,
+    "flash_attention_2": FlashAttention2
+}
+
+class SelfAttention(torch.nn.Module):
+    """Parallel self-attention layer abstract class.
+
+    Self-attention layer takes input with size [s, b, h]
+    and returns output of the same size.
+    """
+
+    def __init__(self, config: ChatGLMConfig, layer_number, device=None):
+        super(SelfAttention, self).__init__()
+        self.layer_number = max(1, layer_number)
+
+        self.projection_size = config.kv_channels * config.num_attention_heads
+
+        # Per attention head and per partition values.
+        self.hidden_size_per_attention_head = self.projection_size // config.num_attention_heads
+        self.num_attention_heads_per_partition = config.num_attention_heads
+
+        self.multi_query_attention = config.multi_query_attention
+        self.qkv_hidden_size = 3 * self.projection_size
+        self.original_rope = config.original_rope
+        if self.multi_query_attention:
+            self.num_multi_query_groups_per_partition = config.multi_query_group_num
+            self.qkv_hidden_size = (
+                    self.projection_size + 2 * self.hidden_size_per_attention_head * config.multi_query_group_num
+            )
+        self.query_key_value = nn.Linear(config.hidden_size, self.qkv_hidden_size,
+                                         bias=config.add_bias_linear or config.add_qkv_bias,
+                                         device=device, **_config_to_kwargs(config)
+                                         )
+
+        self.core_attention = CoreAttention(config, self.layer_number)
+
+        # Output.
+        self.dense = nn.Linear(self.projection_size, config.hidden_size, bias=config.add_bias_linear,
+                               device=device, **_config_to_kwargs(config)
+                               )
+
+    def _allocate_memory(self, inference_max_sequence_len, batch_size, device=None, dtype=None):
+        if self.multi_query_attention:
+            num_attention_heads = self.num_multi_query_groups_per_partition
+        else:
+            num_attention_heads = self.num_attention_heads_per_partition
+        return torch.empty(
+            inference_max_sequence_len,
+            batch_size,
+            num_attention_heads,
+            self.hidden_size_per_attention_head,
+            dtype=dtype,
+            device=device,
+        )
+
+    def forward(
+            self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True
+    ):
+        # hidden_states: [b, sq, h]
+
+        # =================================================
+        # Pre-allocate memory for key-values for inference.
+        # =================================================
+        # =====================
+        # Query, Key, and Value
+        # =====================
+
+        # Attention heads [b, sq, h] --> [b, sq, (np * 3 * hn)]
+        mixed_x_layer = self.query_key_value(hidden_states)
+
+        if self.multi_query_attention:
+            (query_layer, key_layer, value_layer) = mixed_x_layer.split(
+                [
+                    self.num_attention_heads_per_partition * self.hidden_size_per_attention_head,
+                    self.num_multi_query_groups_per_partition * self.hidden_size_per_attention_head,
+                    self.num_multi_query_groups_per_partition * self.hidden_size_per_attention_head,
+                ],
+                dim=-1,
+            )
+            query_layer = query_layer.view(
+                query_layer.size()[:-1] + (self.num_attention_heads_per_partition, self.hidden_size_per_attention_head)
+            )
+            key_layer = key_layer.view(
+                key_layer.size()[:-1] + (self.num_multi_query_groups_per_partition, self.hidden_size_per_attention_head)
+            )
+            value_layer = value_layer.view(
+                value_layer.size()[:-1]
+                + (self.num_multi_query_groups_per_partition, self.hidden_size_per_attention_head)
+            )
+        else:
+            new_tensor_shape = mixed_x_layer.size()[:-1] + \
+                               (self.num_attention_heads_per_partition,
+                                3 * self.hidden_size_per_attention_head)
+            mixed_x_layer = mixed_x_layer.view(*new_tensor_shape)
+
+            # [b, sq, np, 3 * hn] --> 3 [b, sq, np, hn]
+            (query_layer, key_layer, value_layer) = split_tensor_along_last_dim(mixed_x_layer, 3)
+
+        # [b, sq, np, hn] -> [b, np, sq, hn]
+        query_layer, key_layer, value_layer = [k.transpose(1, 2) for k in [query_layer, key_layer, value_layer]]
+
+        # apply relative positional encoding (rotary embedding)
+        if rotary_pos_emb is not None:
+            query_layer = apply_rotary_pos_emb(query_layer, rotary_pos_emb)
+            key_layer = apply_rotary_pos_emb(key_layer, rotary_pos_emb)
+
+        # adjust key and value for inference
+        if kv_cache is not None:
+            cache_k, cache_v = kv_cache
+            key_layer = torch.cat((cache_k, key_layer), dim=2)
+            value_layer = torch.cat((cache_v, value_layer), dim=2)
+        if use_cache:
+            kv_cache = (key_layer, value_layer)
+        else:
+            kv_cache = None
+
+        if self.multi_query_attention:
+            key_layer = key_layer.unsqueeze(2)
+            key_layer = key_layer.expand(
+                -1, -1, self.num_attention_heads_per_partition // self.num_multi_query_groups_per_partition, -1, -1
+            )
+            key_layer = key_layer.contiguous().view(
+                key_layer.size()[:1] + (self.num_attention_heads_per_partition,) + key_layer.size()[3:]
+            )
+            value_layer = value_layer.unsqueeze(2)
+            value_layer = value_layer.expand(
+                -1, -1, self.num_attention_heads_per_partition // self.num_multi_query_groups_per_partition, -1, -1
+            )
+            value_layer = value_layer.contiguous().view(
+                value_layer.size()[:1] + (self.num_attention_heads_per_partition,) + value_layer.size()[3:]
+            )
+
+        # ==================================
+        # core attention computation
+        # ==================================
+
+        context_layer = self.core_attention(query_layer, key_layer, value_layer, attention_mask)
+
+        # =================
+        # Output. [sq, b, h]
+        # =================
+
+        output = self.dense(context_layer)
+
+        return output, kv_cache
+
+
+def _config_to_kwargs(args):
+    common_kwargs = {
+        "dtype": args.torch_dtype,
+    }
+    return common_kwargs
+
+
+class MLP(torch.nn.Module):
+    """MLP.
+
+    MLP will take the input with h hidden state, project it to 4*h
+    hidden dimension, perform nonlinear transformation, and project the
+    state back into h hidden dimension.
+    """
+
+    def __init__(self, config: ChatGLMConfig, device=None):
+        super(MLP, self).__init__()
+
+        self.add_bias = config.add_bias_linear
+
+        # Project to 4h. If using swiglu double the output width, see https://arxiv.org/pdf/2002.05202.pdf
+        self.dense_h_to_4h = nn.Linear(
+            config.hidden_size,
+            config.ffn_hidden_size * 2,
+            bias=self.add_bias,
+            device=device,
+            **_config_to_kwargs(config)
+        )
+
+        def swiglu(x):
+            x = torch.chunk(x, 2, dim=-1)
+            return F.silu(x[0]) * x[1]
+
+        self.activation_func = swiglu
+
+        # Project back to h.
+        self.dense_4h_to_h = nn.Linear(
+            config.ffn_hidden_size,
+            config.hidden_size,
+            bias=self.add_bias,
+            device=device,
+            **_config_to_kwargs(config)
+        )
+
+    def forward(self, hidden_states):
+        # [s, b, 4hp]
+        intermediate_parallel = self.dense_h_to_4h(hidden_states)
+        intermediate_parallel = self.activation_func(intermediate_parallel)
+        # [s, b, h]
+        output = self.dense_4h_to_h(intermediate_parallel)
+        return output
+
+
+class GLMBlock(torch.nn.Module):
+    """A single transformer layer.
+
+    Transformer layer takes input with size [s, b, h] and returns an
+    output of the same size.
+    """
+
+    def __init__(self, config: ChatGLMConfig, layer_number, device=None):
+        super(GLMBlock, self).__init__()
+        self.layer_number = layer_number
+
+        self.apply_residual_connection_post_layernorm = config.apply_residual_connection_post_layernorm
+
+        self.fp32_residual_connection = config.fp32_residual_connection
+
+        LayerNormFunc = RMSNorm if config.rmsnorm else LayerNorm
+        # Layernorm on the input data.
+        self.input_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
+                                             dtype=config.torch_dtype)
+
+        # Self attention.
+        self.self_attention = SelfAttention(config, layer_number, device=device)
+        self.hidden_dropout = config.hidden_dropout
+
+        # Layernorm on the attention output
+        self.post_attention_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
+                                                      dtype=config.torch_dtype)
+
+        # MLP
+        self.mlp = MLP(config, device=device)
+
+    def forward(
+            self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True,
+    ):
+        # hidden_states: [s, b, h]
+
+        # Layer norm at the beginning of the transformer layer.
+        layernorm_output = self.input_layernorm(hidden_states)
+        # Self attention.
+        attention_output, kv_cache = self.self_attention(
+            layernorm_output,
+            attention_mask,
+            rotary_pos_emb,
+            kv_cache=kv_cache,
+            use_cache=use_cache
+        )
+
+        # Residual connection.
+        if self.apply_residual_connection_post_layernorm:
+            residual = layernorm_output
+        else:
+            residual = hidden_states
+
+        layernorm_input = torch.nn.functional.dropout(attention_output, p=self.hidden_dropout, training=self.training)
+        layernorm_input = residual + layernorm_input
+
+        # Layer norm post the self attention.
+        layernorm_output = self.post_attention_layernorm(layernorm_input)
+
+        # MLP.
+        mlp_output = self.mlp(layernorm_output)
+
+        # Second residual connection.
+        if self.apply_residual_connection_post_layernorm:
+            residual = layernorm_output
+        else:
+            residual = layernorm_input
+
+        output = torch.nn.functional.dropout(mlp_output, p=self.hidden_dropout, training=self.training)
+        output = residual + output
+
+        return output, kv_cache
+
+
+class GLMTransformer(torch.nn.Module):
+    """Transformer class."""
+
+    def __init__(self, config: ChatGLMConfig, device=None):
+        super(GLMTransformer, self).__init__()
+
+        self.fp32_residual_connection = config.fp32_residual_connection
+        self.post_layer_norm = config.post_layer_norm
+
+        # Number of layers.
+        self.num_layers = config.num_layers
+
+        # Transformer layers.
+        def build_layer(layer_number):
+            return GLMBlock(config, layer_number, device=device)
+
+        self.layers = torch.nn.ModuleList([build_layer(i + 1) for i in range(self.num_layers)])
+
+        if self.post_layer_norm:
+            LayerNormFunc = RMSNorm if config.rmsnorm else LayerNorm
+            # Final layer norm before output.
+            self.final_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
+                                                 dtype=config.torch_dtype)
+
+        self.gradient_checkpointing = False
+
+    def _get_layer(self, layer_number):
+        return self.layers[layer_number]
+
+    def forward(
+            self, hidden_states, attention_mask, rotary_pos_emb, kv_caches=None,
+            use_cache: Optional[bool] = True,
+            output_hidden_states: Optional[bool] = False,
+    ):
+        if not kv_caches:
+            kv_caches = [None for _ in range(self.num_layers)]
+        presents = () if use_cache else None
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        all_self_attentions = None
+        all_hidden_states = () if output_hidden_states else None
+        for index in range(self.num_layers):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer = self._get_layer(index)
+            if self.gradient_checkpointing and self.training:
+                layer_ret = torch.utils.checkpoint.checkpoint(
+                    layer,
+                    hidden_states,
+                    attention_mask,
+                    rotary_pos_emb,
+                    kv_caches[index],
+                    use_cache,
+                    use_reentrant=False
+                )
+            else:
+                layer_ret = layer(
+                    hidden_states,
+                    attention_mask,
+                    rotary_pos_emb,
+                    kv_cache=kv_caches[index],
+                    use_cache=use_cache
+                )
+            hidden_states, kv_cache = layer_ret
+            if use_cache:
+                presents = presents + (kv_cache,)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        # Final layer norm.
+        if self.post_layer_norm:
+            hidden_states = self.final_layernorm(hidden_states)
+
+        return hidden_states, presents, all_hidden_states, all_self_attentions
+
+
+class ChatGLMPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and
+    a simple interface for downloading and loading pretrained models.
+    """
+
+    is_parallelizable = False
+    supports_gradient_checkpointing = True
+    config_class = ChatGLMConfig
+    base_model_prefix = "transformer"
+    _no_split_modules = ["GLMBlock"]
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+
+    def _init_weights(self, module: nn.Module):
+        """Initialize the weights."""
+        return
+
+    def get_masks(self, input_embeds, past_key_values, padding_mask=None):
+        batch_size, seq_length, embed_size = input_embeds.shape
+        full_attention_mask = torch.ones(batch_size, seq_length, seq_length, device=input_embeds.device)
+        full_attention_mask.tril_()
+        past_length = 0
+        if past_key_values:
+            past_length = past_key_values[0][0].shape[2]
+        if past_length:
+            full_attention_mask = torch.cat((torch.ones(batch_size, seq_length, past_length,
+                                                        device=input_embeds.device), full_attention_mask), dim=-1)
+        if padding_mask is not None:
+            full_attention_mask = full_attention_mask * padding_mask.unsqueeze(1)
+        if not past_length and padding_mask is not None:
+            full_attention_mask -= padding_mask.unsqueeze(-1) - 1
+        full_attention_mask = (full_attention_mask < 0.5).bool()
+        full_attention_mask.unsqueeze_(1)
+        return full_attention_mask
+
+    def get_position_ids(self, input_ids, device):
+        batch_size, seq_length = input_ids.shape
+        position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
+        return position_ids
+
+    def get_multimodal_position_ids(self, input_ids, device):
+        batch_size, seq_length = input_ids.shape
+        position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
+
+class Embedding(torch.nn.Module):
+    """Language model embeddings."""
+
+    def __init__(self, config: ChatGLMConfig, device=None):
+        super(Embedding, self).__init__()
+
+        self.hidden_size = config.hidden_size
+        # Word embeddings (parallel).
+        self.word_embeddings = nn.Embedding(
+            config.padded_vocab_size,
+            self.hidden_size,
+            dtype=config.torch_dtype,
+            device=device
+        )
+        self.fp32_residual_connection = config.fp32_residual_connection
+
+    def forward(self, input_ids):
+        # Embeddings.
+        words_embeddings = self.word_embeddings(input_ids)
+        embeddings = words_embeddings
+        # If the input flag for fp32 residual connection is set, convert for float.
+        if self.fp32_residual_connection:
+            embeddings = embeddings.float()
+        return embeddings
+
+
+def is_empty(images_list: Optional[List[List[torch.Tensor]]]):
+    if images_list is None or len(images_list) == 0:
+        return True
+    for image_list in images_list:
+        if image_list is not None:
+            return False
+    return True
+
+
+class ChatGLMModel(ChatGLMPreTrainedModel):
+    def __init__(self, config: ChatGLMConfig, device=None, empty_init=True):
+        super().__init__(config)
+        if empty_init:
+            init_method = skip_init
+        else:
+            init_method = default_init
+        init_kwargs = {}
+        if device is not None:
+            init_kwargs["device"] = device
+        self.embedding = init_method(Embedding, config, **init_kwargs)
+        self.num_layers = config.num_layers
+        self.multi_query_group_num = config.multi_query_group_num
+        self.kv_channels = config.kv_channels
+
+        # Rotary positional embeddings
+        self.seq_length = config.seq_length
+        rotary_dim = (
+            config.hidden_size // config.num_attention_heads if config.kv_channels is None else config.kv_channels
+        )
+
+        self.rotary_pos_emb = RotaryEmbedding(rotary_dim // 2, rope_ratio=config.rope_ratio,
+                                              original_impl=config.original_rope,
+                                              device=device, dtype=config.torch_dtype)
+        self.encoder = init_method(GLMTransformer, config, **init_kwargs)
+        self.output_layer = init_method(nn.Linear, config.hidden_size, config.padded_vocab_size, bias=False,
+                                        dtype=config.torch_dtype, **init_kwargs)
+        self.pre_seq_len = config.pre_seq_len
+        self.prefix_projection = config.prefix_projection
+        if self.pre_seq_len is not None:
+            for param in self.parameters():
+                param.requires_grad = False
+            self.prefix_tokens = torch.arange(self.pre_seq_len).long()
+            self.prefix_encoder = PrefixEncoder(config)
+            self.dropout = torch.nn.Dropout(0.1)
+
+        self.vision = EVA2CLIPModel(config)
+
+    def get_input_embeddings(self):
+        return self.embedding.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embedding.word_embeddings = value
+
+    def get_prompt(self, batch_size, device, dtype=torch.half):
+        prefix_tokens = self.prefix_tokens.unsqueeze(0).expand(batch_size, -1).to(device)
+        past_key_values = self.prefix_encoder(prefix_tokens).type(dtype)
+        past_key_values = past_key_values.view(
+            batch_size,
+            self.pre_seq_len,
+            self.pre_seq_len,
+            self.num_layers * 2,
+            self.multi_query_group_num,
+            self.kv_channels
+        )
+        # seq_len, b, nh, hidden_size
+        past_key_values = self.dropout(past_key_values)
+        past_key_values = past_key_values.permute([2, 1, 0, 3, 4]).split(2)
+        return past_key_values
+
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            images: torch.Tensor = None,
+            position_ids: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.BoolTensor] = None,
+            full_attention_mask: Optional[torch.BoolTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        """take care of image_encode, position_ids and (attention_mask = None is fine)"""
+
+        # generate mode with past_key_values. the image features are already mapped
+        if past_key_values is None:
+            # not allow for inputs_embeds, because we want to process image feature
+            assert input_ids is not None and inputs_embeds is None, f"{input_ids} {inputs_embeds}"
+            if not is_empty(images):  # multi-modality
+                image_size: int = self.config.vision_config['image_size']
+                patch_size: int = self.config.vision_config['patch_size']
+                num_patches = (image_size // patch_size // 2) ** 2
+                assert len(input_ids) == len(images), f"{len(input_ids)} {len(images)}"
+                inputs_embeds = self.embedding(input_ids)
+
+                images = images.to(dtype=inputs_embeds.dtype)
+                images_features = self.vision(images)
+
+                if position_ids is None:
+                    position_ids = self.get_position_ids(input_ids, device=inputs_embeds.device)
+                new_input_embeds, new_position_ids = [], []
+
+                for i in range(len(input_ids)):
+                    input_id = input_ids[i].tolist()
+                    boi_token_pos, eoi_token_pos = input_id.index(self.config.boi_token_id), input_id.index(
+                        self.config.eoi_token_id)
+                    assert eoi_token_pos - boi_token_pos == 2
+                    new_input_embeds.append(torch.cat(
+                        (inputs_embeds[i, :boi_token_pos], images_features[i].to(inputs_embeds.device),
+                         inputs_embeds[i, eoi_token_pos + 1:])))
+                    new_position_ids.append(torch.cat(
+                        (position_ids[i, :boi_token_pos + 1], position_ids[i, boi_token_pos + 1].repeat(num_patches),
+                         position_ids[i, eoi_token_pos:])
+                    ))
+                inputs_embeds = torch.stack(new_input_embeds, dim=0)
+                position_ids = torch.stack(new_position_ids, dim=0)
+
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        batch_size, seq_length = input_ids.shape
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embedding(input_ids)
+
+        if self.pre_seq_len is not None:
+            if past_key_values is None:
+                past_key_values = self.get_prompt(batch_size=batch_size, device=input_ids.device,
+                                                  dtype=inputs_embeds.dtype)
+            if attention_mask is not None:
+                attention_mask = torch.cat([attention_mask.new_ones((batch_size, self.pre_seq_len)),
+                                            attention_mask], dim=-1)
+
+        if full_attention_mask is None:
+            if (attention_mask is not None and not attention_mask.all()) or (past_key_values and seq_length != 1):
+                if self.training:
+                    # https://github.com/THUDM/GLM-4/issues/264
+                    new_input_ids, new_attention_mask = [], []
+                    for i in range(len(input_ids)):
+                        input_id = input_ids[i].tolist()
+                        boi_token_pos, eoi_token_pos = input_id.index(self.config.boi_token_id), input_id.index(self.config.eoi_token_id)
+                        assert eoi_token_pos - boi_token_pos == 2
+
+                        new_attention_mask.append(torch.cat(
+                            (attention_mask[i, :boi_token_pos + 1], torch.ones(num_patches).to(attention_mask.device),
+                             attention_mask[i, eoi_token_pos:])))
+
+                        new_input_ids.append(torch.cat(
+                            (input_ids[i, :boi_token_pos + 1], input_ids[i, -1].repeat(num_patches),
+                             input_ids[i, eoi_token_pos:])))
+
+                    attention_mask = torch.stack(new_attention_mask, dim=0)
+                    input_ids = torch.stack(new_input_ids, dim=0)
+                    inputs_embeds = self.embedding(input_ids)
+
+                full_attention_mask = self.get_masks(inputs_embeds, past_key_values, padding_mask=attention_mask)
+
+        # Rotary positional embeddings
+        rotary_pos_emb = self.rotary_pos_emb(self.seq_length)
+
+        if position_ids is not None:
+            rotary_pos_emb = rotary_pos_emb[position_ids]
+        else:
+            rotary_pos_emb = rotary_pos_emb[None, :seq_length]
+
+        # Run encoder.
+        hidden_states, presents, all_hidden_states, all_self_attentions = self.encoder(
+            inputs_embeds, full_attention_mask, rotary_pos_emb=rotary_pos_emb,
+            kv_caches=past_key_values, use_cache=use_cache, output_hidden_states=output_hidden_states
+        )
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+def _history_to_prompt(history, query):
+    prompt = ''
+    flag = False
+    for i, (old_query, response) in enumerate(history):
+        prompt += ('<|user|>' if flag else '') + old_query + "<|assistant|>" + response + "<|endoftext|>"
+        flag = True
+    prompt += '{}{}<|assistant|>'.format('<|user|>' if flag else '', query)
+    return prompt
+
+
+class ChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):
+    def __init__(self, config: ChatGLMConfig, empty_init=True, device=None):
+        super().__init__(config)
+
+        self.max_sequence_length = config.max_length
+        self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
+        self.config = config
+
+    def _update_model_kwargs_for_generation(
+            self,
+            outputs: ModelOutput,
+            model_kwargs: Dict[str, Any],
+            is_encoder_decoder: bool = False,
+    ) -> Dict[str, Any]:
+        # update past_key_values
+        cache_name, cache = self._extract_past_from_model_output(outputs)
+        model_kwargs[cache_name] = cache
+
+        # update attention mask
+        if "attention_mask" in model_kwargs:
+            attention_mask = model_kwargs["attention_mask"]
+            model_kwargs["attention_mask"] = torch.cat(
+                [attention_mask, attention_mask.new_ones((attention_mask.shape[0], 1))], dim=-1
+            )
+
+        # update position ids
+        if "position_ids" in model_kwargs:
+            position_ids = model_kwargs["position_ids"]
+            new_position_id = position_ids[..., -1:].clone()
+            new_position_id += 1
+            model_kwargs["position_ids"] = torch.cat(
+                [position_ids, new_position_id], dim=-1
+            )
+
+        model_kwargs["is_first_forward"] = False
+        return model_kwargs
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids: torch.LongTensor,
+            images: Optional[torch.Tensor] = None,
+            past_key_values: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            is_first_forward: bool = True,
+            **kwargs
+    ) -> dict:
+        # only last token for input_ids if past is not None
+        if position_ids is None:
+            position_ids = self.get_position_ids(input_ids, device=input_ids.device)
+        if attention_mask is not None:
+            image_size: int = self.config.vision_config['image_size']
+            patch_size: int = self.config.vision_config['patch_size']
+            num_patches = (image_size // patch_size // 2) ** 2
+            new_attention_masks = []
+
+            # if not image, use this default id
+            eoi_token_pos = 6
+            boi_token_pos = 4
+
+            for i in range(len(input_ids)):
+                input_id = input_ids[i].tolist()
+                if not is_empty(images):
+                    boi_token_pos, eoi_token_pos = input_id.index(self.config.boi_token_id), input_id.index(
+                        self.config.eoi_token_id)
+                assert eoi_token_pos - boi_token_pos == 2
+                new_attention_masks.append(torch.cat(
+                    (attention_mask[i, :boi_token_pos + 1], attention_mask.new_ones(num_patches),
+                     attention_mask[i, eoi_token_pos:])
+                ))
+            attention_mask = torch.stack(new_attention_masks, dim=0)
+        if not is_first_forward:
+            if past_key_values is not None:
+                position_ids = position_ids[..., -1:]
+                input_ids = input_ids[:, -1:]
+        return {
+            "input_ids": input_ids,
+            "images": images,
+            "past_key_values": past_key_values,
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "return_last_logit": True,
+            "use_cache": use_cache
+        }
+
+    def forward(
+            self,
+            input_ids: Optional[torch.Tensor] = None,
+            images: List[List[torch.Tensor]] = None,
+            position_ids: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            labels: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            return_last_logit: Optional[bool] = False,
+    ):
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids=input_ids,
+            images=images,
+            position_ids=position_ids,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = transformer_outputs[0]
+        if return_last_logit:
+            hidden_states = hidden_states[:, -1:]
+        lm_logits = self.transformer.output_layer(hidden_states)
+
+        loss = None
+        if labels is not None:
+            new_labels = []
+            for i in range(len(input_ids)):
+                input_id = input_ids[i].tolist()
+                boi_token_pos, eoi_token_pos = input_id.index(self.config.boi_token_id), input_id.index(
+                    self.config.eoi_token_id)
+                assert eoi_token_pos - boi_token_pos == 2
+
+                new_labels.append(torch.cat(
+                    (
+                        labels[i, :boi_token_pos + 1],
+                        torch.tensor([-100]).to(labels.device).to(labels.dtype).repeat(1600),
+                        labels[i, eoi_token_pos:])))
+
+            labels = torch.stack(new_labels, dim=0)
+            lm_logits = lm_logits.to(torch.float32)
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            loss_fct = CrossEntropyLoss(ignore_index=-100)
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            lm_logits = lm_logits.to(hidden_states.dtype)
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+            past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
+    ) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
+        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+
+        Output shares the same memory storage as `past`.
+        """
+        return tuple(
+            (
+                layer_past[0].index_select(0, beam_idx.to(layer_past[0].device)),
+                layer_past[1].index_select(0, beam_idx.to(layer_past[1].device)),
+            )
+            for layer_past in past
+        )
+
+class ChatGLMForSequenceClassification(ChatGLMPreTrainedModel):
+    def __init__(self, config: ChatGLMConfig, empty_init=True, device=None):
+        super().__init__(config)
+
+        self.num_labels = config.num_labels
+        self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
+
+        self.classifier_head = nn.Linear(config.hidden_size, config.num_labels, bias=True, dtype=torch.half)
+        if config.classifier_dropout is not None:
+            self.dropout = nn.Dropout(config.classifier_dropout)
+        else:
+            self.dropout = None
+        self.config = config
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            full_attention_mask: Optional[torch.Tensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            inputs_embeds: Optional[torch.LongTensor] = None,
+            labels: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor, ...], SequenceClassifierOutputWithPast]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            attention_mask=attention_mask,
+            full_attention_mask=full_attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = transformer_outputs[0]
+        pooled_hidden_states = hidden_states[-1]
+        if self.dropout is not None:
+            pooled_hidden_states = self.dropout(pooled_hidden_states)
+        logits = self.classifier_head(pooled_hidden_states)
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze().float(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits.float(), labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels).float(), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits.float(), labels.view(-1, self.num_labels))
+
+        if not return_dict:
+            output = (logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )

quantization_config.json

@@ -0,0 +1,24 @@
+{
+  "bits": 4,
+  "group_size": 128,
+  "sym": true,
+  "data_type": "int",
+  "enable_quanted_input": true,
+  "enable_minmax_tuning": true,
+  "seqlen": 2048,
+  "batch_size": 8,
+  "scale_dtype": "torch.float16",
+  "lr": 0.001,
+  "minmax_lr": 0.001,
+  "gradient_accumulate_steps": 1,
+  "iters": 1000,
+  "amp": true,
+  "nsamples": 512,
+  "low_gpu_mem_usage": false,
+  "to_quant_block_names": "transformer.encoder.layers",
+  "enable_norm_bias_tuning": false,
+  "dataset": "NeelNanda/pile-10k",
+  "autoround_version": "0.4.2.dev",
+  "quant_method": "intel/auto-round",
+  "backend": "auto_round:gptq:exllamav2"
+}

visual.py

@@ -0,0 +1,180 @@
+import torch
+from torch import nn
+from argparse import Namespace
+import torch.nn.functional as F
+from transformers.activations import ACT2FN
+import math
+from torch.nn import LayerNorm
+
+
+def standard_attention(query_layer, key_layer, value_layer, scaling_attention_score=True):
+    if scaling_attention_score:
+        query_layer = query_layer / math.sqrt(query_layer.shape[-1])
+    attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+    attention_probs = F.softmax(attention_scores, dim=-1)
+
+    context_layer = torch.matmul(attention_probs, value_layer)
+    return context_layer
+
+
+def attention_fn_default(query_layer, key_layer, value_layer, scaling_attention_score=True):
+    if int(torch.__version__.split('.')[0]) >= 2 and scaling_attention_score:
+        # Pytorch 2.0 attention uses very much memory if attention_mask is float, and has NaN bug if attention_mask is None.
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_layer, key_layer, value_layer,
+            attn_mask=None,
+            dropout_p=0.,
+            is_causal=False
+        )
+        return attn_output
+    else:
+        return standard_attention(
+            query_layer, key_layer, value_layer, scaling_attention_score=scaling_attention_score
+        )
+
+
+class PatchEmbedding(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.proj = nn.Conv2d(config.in_channels, config.hidden_size, kernel_size=config.patch_size,
+                              stride=config.patch_size)
+        self.cls_embedding = nn.Parameter(torch.zeros(1, config.hidden_size))
+        self.position_embedding = nn.Embedding(config.num_positions, config.hidden_size)
+
+    def forward(self, images: "tensor(B, C, H, W)") -> "tensor(B, L, D)":
+        x = self.proj(images)
+        x = x.flatten(2).transpose(1, 2)
+        cls_token = self.cls_embedding.expand(x.shape[0], -1, -1)
+        x = torch.cat((cls_token, x), dim=1)
+        x += self.position_embedding.weight.unsqueeze(0)
+        return x
+
+
+class Attention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.num_heads = config.num_heads
+        head_dim = config.hidden_size // config.num_heads
+        self.scale = head_dim ** -0.5
+        self.query_key_value = nn.Linear(config.hidden_size, config.hidden_size * 3)
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.output_dropout = torch.nn.Dropout(config.dropout_prob)
+
+    def forward(self, x: "tensor(B, L, D)") -> "tensor(B, L, D)":
+        B, L, _ = x.shape
+        qkv = self.query_key_value(x)
+        qkv = qkv.reshape(B, L, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)  # 3, B, H, L, D
+        q, k, v = qkv[0], qkv[1], qkv[2]
+
+        out = attention_fn_default(
+            q, k, v
+        )
+        output = self.dense(out.transpose(1, 2).reshape(B, L, -1))
+        output = self.output_dropout(output)
+        return output
+
+    def attention(self, q, k, v):
+        attn_weights = torch.matmul(q * self.scale, k.transpose(-2, -1))
+        attn_weights = attn_weights.softmax(dim=-1)
+        output = torch.matmul(attn_weights, v)
+        return output
+
+
+class MLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.activation_fn = ACT2FN[config.hidden_act]
+        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
+        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.fc1(x)
+        x = self.activation_fn(x)
+        x = self.fc2(x)
+        return x
+
+
+class TransformerLayer(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.input_layernorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.attention = Attention(config)
+        self.mlp = MLP(config)
+        self.post_attention_layernorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        attention_input = hidden_states
+        attention_output = self.input_layernorm(self.attention(attention_input))
+        hidden_states = attention_input + attention_output
+        mlp_input = hidden_states
+
+        # https://github.com/THUDM/GLM-4/issues/350
+        mlp_output = self.post_attention_layernorm(self.mlp(mlp_input)).to(mlp_input.device)
+        output = mlp_input + mlp_output
+        return output
+
+
+class Transformer(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.layers = nn.ModuleList([TransformerLayer(config) for _ in range(config.num_hidden_layers)])
+
+    def forward(self, hidden_states):
+        for layer_module in self.layers:
+            hidden_states = layer_module(hidden_states)
+        return hidden_states
+
+
+class GLU(nn.Module):
+    def __init__(self, config, in_features):
+        super().__init__()
+        self.linear_proj = nn.Linear(in_features, config.hidden_size, bias=False)
+        self.norm1 = nn.LayerNorm(config.hidden_size)
+        self.act1 = nn.GELU()
+        self.act2 = nn.functional.silu
+        self.dense_h_to_4h = nn.Linear(config.hidden_size, config.ffn_hidden_size, bias=False)
+        self.gate_proj = nn.Linear(config.hidden_size, config.ffn_hidden_size, bias=False)
+        self.dense_4h_to_h = nn.Linear(config.ffn_hidden_size, config.hidden_size, bias=False)
+
+    def forward(self, x):
+        x = self.linear_proj(x)
+        x = self.act1(self.norm1(x))
+        x = self.act2(self.gate_proj(x)) * self.dense_h_to_4h(x)
+        x = self.dense_4h_to_h(x)
+        return x
+
+
+class EVA2CLIPModel(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        vision_config = Namespace(**config.vision_config)
+        self.patch_embedding = PatchEmbedding(vision_config)
+        self.transformer = Transformer(vision_config)
+        self.linear_proj = GLU(config, in_features=config.hidden_size)
+        self.conv = nn.Conv2d(in_channels=vision_config.hidden_size, out_channels=config.hidden_size, kernel_size=2,
+                              stride=2)
+        self.boi = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
+        self.eoi = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
+        self.scaling_factor = vision_config.scaling_factor
+
+    def forward(self, images: "tensor(B, C, H, W)") -> "tensor(B, L, D)":
+        x = self.patch_embedding(images)
+        x = self.transformer(x)
+        x = x[:, 1:]
+
+        b, s, h = x.shape
+        grid_size = int(s ** 0.5)
+        x = x.view(b, grid_size, grid_size, h).permute(0, 3, 1, 2)
+        x = self.conv(x)
+
+        x = x.flatten(2).transpose(1, 2)
+        x = self.linear_proj(x)
+
+        # https://github.com/THUDM/GLM-4/issues/350
+        boi = self.boi.expand(x.shape[0], -1, -1).to(x.device)
+        eoi = self.eoi.expand(x.shape[0], -1, -1).to(x.device)
+        x = torch.cat((boi, x, eoi), dim=1)
+        x = x / self.scaling_factor
+        return x