megatron

config.json

@@ -0,0 +1,51 @@
+{
+    "_from_model_config": true,
+  "architectures": [
+    "YuanForCausalLM"
+  ],
+  "auto_map":{
+          "AutoConfig": "configuration_yuan.YuanConfig",
+          "AutoModelForCausalLM": "yuan_hf_model.YuanForCausalLM"
+  },
+  "tokenizer_class": "YuanTokenizer",
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 4096,
+  "model_type": "yuan",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "rms_norm_eps": 1e-06,
+  "dropout": 0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.30.0.dev0",
+  "use_cache": true,
+  "causal_mask": true,
+  "use_flash_attention": true,
+  "reset_attention_mask": true,
+  "reset_position_ids": true,
+  "use_loss_mask": false,
+  "use_moe": true,
+  "moe_config":{
+      "moe_num_experts": 32,
+      "moe_top_k": 2,
+      "ffn_hidden_size": 8192,
+      "norm_topk_prob":true,
+      "gated_linear_unit":true
+  },
+  "eod_token": 77185,
+  "sep_token": 77187,
+  "eod_token_id": 77185,
+  "sep_token_id": 77185,
+  "pad_token_id": 77185,
+  "bos_token_id": 77185,
+  "eos_token_id": 77185,
+  "mask_token_id": 77185,
+  "vocab_size": 135040,
+  "attention_projection_size": 4096,
+  "reset_attention_mask": false,
+  "output_router_logits": true
+}
+

configuration_yuan.py

@@ -0,0 +1,44 @@
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class YuanConfig(PretrainedConfig):
+    model_type = "yuan"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=135040,
+        hidden_size=2048,
+        intermediate_size=8192,
+        num_hidden_layers=24,
+        num_attention_heads=32,
+        hidden_act="silu",
+        model_max_length=8192,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=77185,
+        bos_token_id=77185,
+        eos_token_id=77185,
+        tie_word_embeddings=True,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.model_max_length = model_max_length
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+

iter_0001477/mp_rank_00_000/model_optim_rng.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e26d1367a3a5eae075bad849d8d651b8cb4a61f7831b9c376ddf409d7989762
+size 10469760007

iter_0001477/mp_rank_00_001/model_optim_rng.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:067dff491248a5c43434fe329632d0c31d73322e0617549d063183549b3b7c27
+size 9916635714

latest_checkpointed_iteration.txt

@@ -0,0 +1 @@
+1477