Upload checkpoints

README.md

@@ -0,0 +1,12 @@
+---
+inference: false
+---
+
+# Model Card
+
+Bunny is a family of lightweight multimodal models.
+
+Bunny-pretrain-phi-2-eva is the pretrained weights for [bunny-phi-2-eva-lora](https://huggingface.co/BoyaWu10/bunny-phi-2-eva-lora), which leverages Phi-2 as the language model backbone and EVA-CLIP 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 +1,40 @@
+{
+  "_name_or_path": "bunny-pretrain-phi-2-eva",
+  "architectures": [
+    "PhiForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.0,
+  "eos_token_id": 50256,
+  "freeze_mm_mlp_adapter": false,
+  "hidden_act": "gelu_new",
+  "hidden_size": 2560,
+  "image_aspect_ratio": "square",
+  "initializer_range": 0.02,
+  "intermediate_size": 10240,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 2048,
+  "mm_hidden_size": 1024,
+  "mm_projector_lr": null,
+  "mm_projector_type": "mlp2x_gelu",
+  "mm_vision_tower": "/path/to/EVA02_CLIP_L_336_psz14_s6B.pt",
+  "model_type": "bunny-phi",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "partial_rotary_factor": 0.4,
+  "pretraining_tp": 1,
+  "qk_layernorm": false,
+  "resid_pdrop": 0.1,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "tokenizer_model_max_length": 2048,
+  "tokenizer_padding_side": "right",
+  "torch_dtype": "float16",
+  "transformers_version": "4.36.2",
+  "tune_mm_mlp_adapter": true,
+  "use_cache": true,
+  "use_mm_proj": true,
+  "vocab_size": 51200
+}

mm_projector.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b70c419ce6011b66a7c7ed507a5af49b65fb78f5319468c9948c55aa1a58fb8
+size 18362424