Upload checkpoints

README.md

@@ -0,0 +1,12 @@
+---
+inference: false
+---
+
+# Model Card
+
+Bunny is a family of lightweight multimodal models.
+
+Bunny-pretrain-stablelm-2-siglip is the pretrained weights for [bunny-stablelm-2-siglip-lora](https://huggingface.co/BoyaWu10/bunny-stablelm-2-siglip-lora), which leverages StableLM-2 as the language model backbone and SigLIP as the vision encoder.
+It is pretrained on LAION-2M.
+
+More details about this model can be found in [GitHub](https://github.com/BAAI-DCAI/Bunny).

config.json

@@ -0,0 +1,42 @@
+{
+  "_name_or_path": "bunny-pretrain-stablelm-2-siglip",
+  "architectures": [
+    "StableLMEpochForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "bunny_stablelm.BunnyStableLMConfig",
+    "AutoModelForCausalLM": "modeling_stablelm_epoch.StableLMEpochForCausalLM"
+  },
+  "bos_token_id": null,
+  "eos_token_id": 100257,
+  "freeze_mm_mlp_adapter": false,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "image_aspect_ratio": "square",
+  "initializer_range": 0.02,
+  "intermediate_size": 5632,
+  "max_position_embeddings": 4096,
+  "mm_hidden_size": 1152,
+  "mm_projector_lr": null,
+  "mm_projector_type": "mlp2x_gelu",
+  "mm_vision_tower": "google/siglip-so400m-patch14-384",
+  "model_type": "bunny-stablelm",
+  "norm_eps": 1e-05,
+  "num_attention_heads": 32,
+  "num_heads": 32,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 32,
+  "rope_pct": 0.25,
+  "rope_theta": 10000,
+  "rotary_scaling_factor": 1.0,
+  "tie_word_embeddings": false,
+  "tokenizer_model_max_length": 2048,
+  "tokenizer_padding_side": "right",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.36.2",
+  "tune_mm_mlp_adapter": true,
+  "use_cache": true,
+  "use_mm_proj": true,
+  "use_qkv_bias": true,
+  "vocab_size": 100352
+}

mm_projector.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:537299b5730f459f84f325c55ee98a8170acdcd2530077692640349a69e39127
+size 13117496