Upload folder using huggingface_hub

config.json

@@ -0,0 +1,73 @@
+{
+  "architectures": [
+    "OpenCUAForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_opencua.OpenCUAConfig",
+    "AutoModel": "modeling_opencua.OpenCUAForConditionalGeneration",
+    "AutoModelForCausalLM": "modeling_opencua.OpenCUAForConditionalGeneration",
+    "AutoProcessor": "processing_opencua.OpenCUAProcessor"
+  },
+  "ignore_index": -100,
+  "media_placeholder_token_id": 151664,
+  "model_type": "opencua",
+  "pad_token_id": 0,
+  "num_attention_heads": 28,
+  "max_position_embeddings": 128000,
+  "text_config": {
+    "bos_token_id": 151643,
+    "eos_token_id": 151644,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 3584,
+    "initializer_range": 0.02,
+    "intermediate_size": 18944,
+    "k_proj_bias": true,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "qwen2",
+    "num_attention_heads": 28,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 28,
+    "num_key_value_heads": 4,
+    "pad_token_id": 152063,
+    "max_position_embeddings": 128000,
+    "pretraining_sequence_length": 128000,
+    "q_proj_bias": true,
+    "rms_norm_eps": 1e-05,
+    "rope_theta": 1000000.0,
+    "tie_word_embeddings": false,
+    "torch_dtype": "bfloat16",
+    "use_bfloat16": false,
+    "use_cache": true,
+    "v_proj_bias": true,
+    "vocab_size": 152064
+  },
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.48.3",
+  "vision_config": {
+    "depth": 32,
+    "fullatt_block_indexes": [
+      7,
+      15,
+      23,
+      31
+    ],
+    "hidden_act": "silu",
+    "hidden_size": 1280,
+    "num_heads": 16,
+    "in_chans": 3,
+    "intermediate_size": 3420,
+    
+    "patch_size": 14,
+    "spatial_merge_size": 2,
+    "spatial_patch_size": 14,
+    "temporal_patch_size": 2,
+    "out_hidden_size": 3584,
+    "tokens_per_second": 2,
+    "window_size": 112
+  },
+  "vocab_size": 152064
+}

configuration_opencua.py

@@ -0,0 +1,38 @@
+from transformers.configuration_utils import PretrainedConfig
+from transformers.models.qwen2_5_vl.configuration_qwen2_5_vl import Qwen2_5_VLVisionConfig
+from transformers.models.qwen2.configuration_qwen2 import Qwen2Config
+
+
+class OpenCUAConfig(PretrainedConfig):
+    """OpenCUA-2.5-7B model configuration.
+
+    Args:
+        vision_config: Configuration for the vision model.Qwen2_5_VLVisionConfig
+        text_config: Configuration for the text model. Qwen2Config
+        pad_token_id: The token ID to use for padding.
+    """
+
+    model_type = "opencua"
+
+    def __init__(
+        self,
+        vision_config: dict | Qwen2_5_VLVisionConfig | None = None,
+        text_config: dict | Qwen2Config | None = None,
+        ignore_index: int = -100,
+        media_placeholder_token_id: int = 151664,
+        pad_token_id: int = 0,
+        **kwargs
+    ):
+        if isinstance(vision_config, dict):
+            vision_config = Qwen2_5_VLVisionConfig(**vision_config)
+        self.vision_config = vision_config
+
+        if isinstance(text_config, dict):
+            text_config = Qwen2Config(**text_config)
+        self.text_config = text_config
+
+        self.ignore_index = ignore_index
+        self.media_placeholder_token_id = media_placeholder_token_id
+
+        super().__init__(pad_token_id=pad_token_id, **kwargs)
+        

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "max_length": 32768,
+  "eos_token_id": [
+     151645,
+     151644,
+     151643
+  ]
+}

model-1-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3d6b25647e2b587c73a4218d486fdcfe871841db79484f07a7923677e525473
+size 466117136

model-10-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adf476ad9901a23367b16985ead74683a3aafc385ecf1d99a6b28ca21cfe703e
+size 466117136

model-11-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a9fa2d9cf00f3f7f821b5b4d54dd7eb6b4084400fdafb8078578f0d07917e22
+size 466117144

model-12-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb0e67d7bbd78437081171464a0c7c57f1549a9090a2fb42dc73b19d94fba495
+size 466117144

model-13-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2dca2c5bf93a7d4563915e8c54bd1c93741180de3faaa8b8834846e253e557fb
+size 466117144

model-14-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e001b216b79438b4a69224bff91c35c4fc6ba9dc78dcda93294f193376f893c
+size 466117144

model-15-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a703f939c1760c207526014edcae70d85f07e9f565f2875a155c4b9299378cf7
+size 466117144

model-16-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfd82bf2141a4fb0e11d79938d1b2ac14369677d9a27e42a30de772fa9f27e5e
+size 466117144

model-17-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35cee47e7c5a95bd5f12616f42038c9afefc4b5630046da6a28147a1f0599d75
+size 466117144

model-18-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:017dddcd4ace4d3e05c4be011033d73c68e1106a15c9ecd3a95ce352e70cd295
+size 466117144

model-19-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02dcdc01ebcdd1c35cbf1d072e1a20eaa82730dd3858a0ab8df6f469b4b6c97f
+size 466117144

model-2-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46203fb402929e58025550b29692df1c3784cac2dab3ba8bf0ab3d12dfd3cb0e
+size 466117136

model-20-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:231f45b7d467015e9777cb32fd6c601221c6fcc37861bbaf051a924212d7ddb1
+size 466117144

model-21-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a21d6b0a76dcc9d5d8a6eb78d37c63417a42ca15da427be916ec8d542ea8a079
+size 466117144

model-22-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc0b4cac4165f88c8730a9f28fb7ea1cf42f6ce7177206968f81df95777eca97
+size 466117144

model-23-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae7247b476ac2b9c13ca49b9fe898625f5a319bb736e9c1257223db296559811
+size 466117144

model-24-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a57eaed9ad51ee020baae852de06fa03670fb4060bb0b40912f50ee31d02260
+size 466117144

model-25-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23bad1d324b717a327481d9c8e10eb3029dc9e92ef578d826c88e7ba6b12e716
+size 466117144

model-26-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae5739717c8ddf2a9d70affe8c06f6f76d7fc4b2c52c9646ab921a0fa73c2668
+size 466117144

model-27-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95c6612dbb6b658237a43f40957c5262800fb9e71c1fc94b0ab0b4fc5e78551c
+size 466117144

model-28-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c88af7b77c1fc7766c6ca6f5aaa1c02071f5de4393981dbddbc74eddd2c1e04b
+size 466117144

model-29-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2709f5ac915b7e5ac5eef2e95c521025d46d2cfd16de34a61f88b3c5b5692f3a
+size 2179997040

model-3-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68c5f3751c6e4c5191c66d1004ef0207a17f1278f925851ea8b59ea5b6d6707a
+size 466117136

model-30-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f01ab69ff0ebeb39e0aed61cff54f76c41a05af328086ecaff487f1bb62c2263
+size 1353144368

model-4-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69a5868b7982fc9a31296cfc995e374ca749dd223139f9058ffd758675da1c4c
+size 466117136

model-5-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f68200835873dc96c23eb15b7e700b6220e91d395650fe2dc69f67232b743470
+size 466117136

model-6-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01b17e6b830001e47f06f1dc3044604a5f930083f36c0f7d54b66a7aa4ba2ba4
+size 466117136

model-7-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d57910cb8778491b512099d2541944c619fe51121d03183e7c1dda401f5425dc
+size 466117136

model-8-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:851747abf0fd9c150c7cfbcae4a6b0d0e2b307f2a421467abf7ed290cc8fabd7
+size 466117136

model-9-of-30.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db02221c8b90a246e0582908af17af6d1f12d6ccce6b7c1c5b6902bc5f7977bf
+size 466117136

model.safetensors.index.json

@@ -0,0 +1,736 @@
+{
+  "metadata": {
+    "total_size": 16584333312
+  },
+  "weight_map": {
+    "language_model.model.layers.19.self_attn.k_proj.bias": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.self_attn.q_proj.bias": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.self_attn.k_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.self_attn.q_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.mlp.up_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.mlp.down_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.mlp.gate_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.post_attention_layernorm.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.input_layernorm.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.self_attn.v_proj.bias": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.self_attn.o_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.19.self_attn.v_proj.weight": "model-20-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.k_proj.bias": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.q_proj.bias": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.k_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.q_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.mlp.up_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.mlp.down_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.mlp.gate_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.post_attention_layernorm.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.input_layernorm.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.v_proj.bias": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.o_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.12.self_attn.v_proj.weight": "model-13-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.k_proj.bias": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.q_proj.bias": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.k_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.q_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.mlp.up_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.mlp.down_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.mlp.gate_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.post_attention_layernorm.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.input_layernorm.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.v_proj.bias": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.o_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.1.self_attn.v_proj.weight": "model-2-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.k_proj.bias": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.q_proj.bias": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.k_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.q_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.mlp.up_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.mlp.down_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.mlp.gate_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.post_attention_layernorm.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.input_layernorm.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.v_proj.bias": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.o_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.26.self_attn.v_proj.weight": "model-27-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.k_proj.bias": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.q_proj.bias": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.k_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.q_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.mlp.up_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.mlp.down_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.mlp.gate_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.post_attention_layernorm.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.input_layernorm.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.v_proj.bias": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.o_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.11.self_attn.v_proj.weight": "model-12-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.k_proj.bias": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.q_proj.bias": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.k_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.q_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.mlp.up_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.mlp.down_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.mlp.gate_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.post_attention_layernorm.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.input_layernorm.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.v_proj.bias": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.o_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.14.self_attn.v_proj.weight": "model-15-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.k_proj.bias": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.q_proj.bias": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.k_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.q_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.mlp.up_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.mlp.down_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.mlp.gate_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.post_attention_layernorm.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.input_layernorm.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.v_proj.bias": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.o_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.23.self_attn.v_proj.weight": "model-24-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.k_proj.bias": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.q_proj.bias": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.k_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.q_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.mlp.up_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.mlp.down_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.mlp.gate_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.post_attention_layernorm.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.input_layernorm.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.v_proj.bias": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.o_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.25.self_attn.v_proj.weight": "model-26-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.k_proj.bias": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.q_proj.bias": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.k_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.q_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.mlp.up_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.mlp.down_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.mlp.gate_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.post_attention_layernorm.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.input_layernorm.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.v_proj.bias": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.o_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.9.self_attn.v_proj.weight": "model-10-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.k_proj.bias": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.q_proj.bias": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.k_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.q_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.mlp.up_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.mlp.down_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.mlp.gate_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.post_attention_layernorm.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.input_layernorm.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.v_proj.bias": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.o_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.20.self_attn.v_proj.weight": "model-21-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.k_proj.bias": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.q_proj.bias": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.k_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.q_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.mlp.up_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.mlp.down_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.mlp.gate_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.post_attention_layernorm.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.input_layernorm.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.v_proj.bias": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.o_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.21.self_attn.v_proj.weight": "model-22-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.k_proj.bias": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.q_proj.bias": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.k_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.q_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.mlp.up_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.mlp.down_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.mlp.gate_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.post_attention_layernorm.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.input_layernorm.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.v_proj.bias": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.o_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.27.self_attn.v_proj.weight": "model-28-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.k_proj.bias": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.q_proj.bias": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.k_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.q_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.mlp.up_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.mlp.down_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.mlp.gate_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.post_attention_layernorm.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.input_layernorm.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.v_proj.bias": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.o_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.18.self_attn.v_proj.weight": "model-19-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.k_proj.bias": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.q_proj.bias": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.k_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.q_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.mlp.up_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.mlp.down_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.mlp.gate_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.post_attention_layernorm.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.input_layernorm.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.v_proj.bias": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.o_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.0.self_attn.v_proj.weight": "model-1-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.k_proj.bias": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.q_proj.bias": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.k_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.q_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.mlp.up_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.mlp.down_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.mlp.gate_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.post_attention_layernorm.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.input_layernorm.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.v_proj.bias": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.o_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.5.self_attn.v_proj.weight": "model-6-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.k_proj.bias": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.q_proj.bias": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.k_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.q_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.mlp.up_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.mlp.down_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.mlp.gate_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.post_attention_layernorm.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.input_layernorm.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.v_proj.bias": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.o_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.6.self_attn.v_proj.weight": "model-7-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.k_proj.bias": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.q_proj.bias": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.k_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.q_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.mlp.up_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.mlp.down_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.mlp.gate_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.post_attention_layernorm.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.input_layernorm.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.v_proj.bias": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.o_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.4.self_attn.v_proj.weight": "model-5-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.k_proj.bias": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.q_proj.bias": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.k_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.q_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.mlp.up_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.mlp.down_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.mlp.gate_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.post_attention_layernorm.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.input_layernorm.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.v_proj.bias": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.o_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.15.self_attn.v_proj.weight": "model-16-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.k_proj.bias": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.q_proj.bias": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.k_proj.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.q_proj.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.mlp.up_proj.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.mlp.down_proj.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.mlp.gate_proj.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.post_attention_layernorm.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.input_layernorm.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.v_proj.bias": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.o_proj.weight": "model-23-of-30.safetensors",
+    "language_model.model.layers.22.self_attn.v_proj.weight": "model-23-of-30.safetensors",
+    "language_model.lm_head.weight": "model-29-of-30.safetensors",
+    "language_model.model.norm.weight": "model-29-of-30.safetensors",
+    "language_model.model.embed_tokens.weight": "model-29-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.k_proj.bias": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.q_proj.bias": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.k_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.q_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.mlp.up_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.mlp.down_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.mlp.gate_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.post_attention_layernorm.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.input_layernorm.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.v_proj.bias": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.o_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.7.self_attn.v_proj.weight": "model-8-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.k_proj.bias": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.q_proj.bias": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.k_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.q_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.mlp.up_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.mlp.down_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.mlp.gate_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.post_attention_layernorm.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.input_layernorm.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.v_proj.bias": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.o_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.16.self_attn.v_proj.weight": "model-17-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.k_proj.bias": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.q_proj.bias": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.k_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.q_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.mlp.up_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.mlp.down_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.mlp.gate_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.post_attention_layernorm.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.input_layernorm.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.v_proj.bias": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.o_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.8.self_attn.v_proj.weight": "model-9-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.k_proj.bias": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.q_proj.bias": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.k_proj.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.q_proj.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.mlp.up_proj.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.mlp.down_proj.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.mlp.gate_proj.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.post_attention_layernorm.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.input_layernorm.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.v_proj.bias": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.o_proj.weight": "model-25-of-30.safetensors",
+    "language_model.model.layers.24.self_attn.v_proj.weight": "model-25-of-30.safetensors",
+    "vision_tower.blocks.0.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.merger.mlp.2.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.merger.mlp.0.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.attn.qkv.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.merger.ln_q.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.norm2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.norm1.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.mlp.down_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.merger.mlp.2.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.mlp.down_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.mlp.gate_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.mlp.up_proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.attn.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.patch_embed.proj.weight": "model-30-of-30.safetensors",
+    "vision_tower.merger.mlp.0.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.attn.qkv.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.9.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.8.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.6.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.5.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.30.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.3.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.29.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.27.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.26.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.24.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.23.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.21.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.20.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.2.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.18.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.17.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.15.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.14.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.12.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.11.mlp.up_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.0.mlp.gate_proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.7.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.4.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.31.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.28.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.25.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.22.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.19.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.16.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.13.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.10.attn.proj.bias": "model-30-of-30.safetensors",
+    "vision_tower.blocks.1.attn.proj.bias": "model-30-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.k_proj.bias": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.q_proj.bias": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.k_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.q_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.mlp.up_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.mlp.down_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.mlp.gate_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.post_attention_layernorm.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.input_layernorm.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.v_proj.bias": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.o_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.3.self_attn.v_proj.weight": "model-4-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.k_proj.bias": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.q_proj.bias": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.k_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.q_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.mlp.up_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.mlp.down_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.mlp.gate_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.post_attention_layernorm.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.input_layernorm.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.v_proj.bias": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.o_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.10.self_attn.v_proj.weight": "model-11-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.k_proj.bias": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.q_proj.bias": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.k_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.q_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.mlp.up_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.mlp.down_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.mlp.gate_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.post_attention_layernorm.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.input_layernorm.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.v_proj.bias": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.o_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.13.self_attn.v_proj.weight": "model-14-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.k_proj.bias": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.q_proj.bias": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.k_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.q_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.mlp.up_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.mlp.down_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.mlp.gate_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.post_attention_layernorm.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.input_layernorm.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.v_proj.bias": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.o_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.2.self_attn.v_proj.weight": "model-3-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.k_proj.bias": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.q_proj.bias": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.k_proj.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.q_proj.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.mlp.up_proj.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.mlp.down_proj.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.mlp.gate_proj.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.post_attention_layernorm.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.input_layernorm.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.v_proj.bias": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.o_proj.weight": "model-18-of-30.safetensors",
+    "language_model.model.layers.17.self_attn.v_proj.weight": "model-18-of-30.safetensors"
+  }
+}

modeling_opencua.py

@@ -0,0 +1,499 @@
+# ------------------------------------------------------------------------------
+# OpenCUA‑7B Model
+#
+# This implementation is adapted from the Qwen2‑VL reference code in
+# Hugging Face Transformers v4.53.0:
+#   https://github.com/huggingface/transformers/tree/v4.53.0/src/transformers/models/qwen2_5_vl
+#
+# Checkpoint used for weight initialisation:
+#   "Qwen/Qwen2.5-VL-7B-Instruct"  –  https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct
+#
+# Key modifications
+# -----------------
+# • Replaced Multimodal Rotary Position Embedding (M‑RoPE) with 1‑D RoPE for
+#   compatibility with OpenCUA training settings.
+# • Wrapped vision encoder and language model into a single
+#   `OpenCUAForConditionalGeneration` class.
+# • Simplified weight initialisation — this file targets inference / fine‑tuning,
+#   not training from scratch.
+#
+# Copyright (c) 2025 XLANG Lab, The University of Hong Kong
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the “Software”), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+# 
+# ------------------------------------------------------------------------------
+# Prohibited Uses & Additional Disclaimer
+# ---------------------------------------
+# • The Software may **not** be used for any purpose or activity that violates
+#   applicable laws or regulations in any jurisdiction.
+# • The authors, contributors, and copyright holders are **not responsible**
+#   for any illegal, unethical, or harmful use of the Software, nor for any
+#   direct or indirect damages resulting from such use.
+# • Use of the “OpenCUA” name, logo, or trademarks does **not** imply any
+#   endorsement or affiliation unless a separate written permission is obtained.
+
+import torch
+import torch.nn as nn
+from transformers.cache_utils import Cache
+from transformers.modeling_utils import PreTrainedModel
+from transformers.models.llava.modeling_llava import LlavaCausalLMOutputWithPast
+
+from .configuration_opencua import OpenCUAConfig
+from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import Qwen2_5_VisionTransformerPretrainedModel
+from transformers.models.qwen2.modeling_qwen2 import Qwen2ForCausalLM
+
+from typing import Optional
+
+class OpenCUAPreTrainedModel(PreTrainedModel):
+    config_class = OpenCUAConfig
+    base_model_prefix = "language_model"
+    _no_split_modules = ["Qwen2_5_VisionTransformerPretrainedModel", "Qwen2DecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+
+    supports_gradient_checkpointing = True
+
+    _supports_sdpa = True
+
+    def _init_weights(self, module):
+        # important: this ported version of Llava isn't meant for training from scratch - only
+        # inference and fine-tuning - so the proper init weights code has been removed - the original codebase
+        # https://github.com/haotian-liu/LLaVA/tree/main/llava should serve for that purpose
+        std = (
+            self.config.initializer_range
+            if hasattr(self.config, "initializer_range")
+            else self.config.text_config.initializer_range
+        )
+
+        if hasattr(module, "class_embedding"):
+            module.class_embedding.data.normal_(mean=0.0, std=std)
+
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+class OpenCUAForConditionalGeneration(OpenCUAPreTrainedModel):
+
+    def __init__(self, config: OpenCUAConfig):
+        super().__init__(config)
+        self.vision_tower = Qwen2_5_VisionTransformerPretrainedModel(config.vision_config)
+        self.language_model = Qwen2ForCausalLM(config.text_config)
+        self._supports_sdpa = True
+        self.post_init()
+    
+    # 使用 property 来创建动态属性
+    @property
+    def model(self):
+        return self.language_model.model
+    
+    @property
+    def lm_head(self):
+        return self.language_model.lm_head
+        
+    def get_input_embeddings(self):
+        return self.language_model.get_input_embeddings()
+
+    def set_input_embeddings(self, value):
+        self.language_model.set_input_embeddings(value)
+
+    def get_output_embeddings(self):
+        return self.language_model.get_output_embeddings()
+
+    def set_output_embeddings(self, new_embeddings):
+        self.language_model.set_output_embeddings(new_embeddings)
+
+    def set_decoder(self, decoder):
+        self.language_model.set_decoder(decoder)
+
+    def get_decoder(self):
+        return self.language_model.get_decoder()
+
+    def tie_weights(self):
+        return self.language_model.tie_weights()
+
+    def resize_token_embeddings(self, new_num_tokens: int | None = None, pad_to_multiple_of=None) -> nn.Embedding:
+        model_embeds = self.language_model.resize_token_embeddings(
+            new_num_tokens, pad_to_multiple_of)
+        # update vocab size
+        self.config.text_config.vocab_size = model_embeds.num_embeddings
+        self.vocab_size = model_embeds.num_embeddings
+        return model_embeds
+
+    def _merge_input_ids_with_image_features(
+            self,
+            image_features: torch.Tensor,
+            feature_lengths: list[int],
+            inputs_embeds: torch.Tensor,
+            input_ids: torch.Tensor,
+            attention_mask: torch.Tensor,
+            labels: torch.Tensor | None = None):
+        """
+        Args:
+            image_features (:obj:`torch.Tensor` of shape :obj:`(num_image_tokens, embed_dim)`):
+                The image features to merge with the input embeddings.
+            feature_lengths: the length of image feature.
+            inputs_embeds (:obj:`torch.Tensor` of shape :obj:`(batch_size, sequence_length, embed_dim)`):
+                The input embeddings.
+            input_ids (:obj:`torch.Tensor` of shape :obj:`(batch_size, sequence_length)`):
+                The input ids.
+            attention_mask (:obj:`torch.Tensor` of shape :obj:`(batch_size, sequence_length)`):
+                The attention mask.
+            labels (:obj:`torch.Tensor` of shape :obj:`(batch_size, sequence_length)`, *optional*):
+                The labels.
+        """
+
+        image_token_index: int = self.config.media_placeholder_token_id
+        pad_token_id: int = self.config.pad_token_id
+        ignore_index: int = self.config.ignore_index
+
+        _, embed_dim = image_features.shape
+
+        batch_size, sequence_length = input_ids.shape
+        left_padding = not torch.sum(
+            input_ids[:, -1] == torch.tensor(pad_token_id))
+
+        # 1. Create a mask to know where special image tokens are
+        _token_occupation_table = torch.ones_like(input_ids.flatten())
+        _token_occupation_table[input_ids.flatten() == image_token_index] = \
+            torch.tensor(feature_lengths,
+                         dtype=torch.long, device=input_ids.device)
+        _token_occupation_table = _token_occupation_table.reshape(
+            input_ids.shape)
+
+        max_embed_dim = _token_occupation_table.sum(-1).max().item()
+        assert max_embed_dim >= sequence_length, (
+            f"The maximum embedding dimension ({max_embed_dim}) is less than the sequence length ({sequence_length})"
+        )
+        batch_indices, non_image_indices = torch.where(input_ids != image_token_index)
+
+        # 2. Compute the positions where text should be written
+        # Calculate new positions for text tokens in merged image-text sequence.
+        new_token_positions = torch.cumsum(_token_occupation_table, -1) - 1
+        nb_image_pad = max_embed_dim - 1 - new_token_positions[:, -1]
+        if left_padding:
+            new_token_positions += nb_image_pad[:, None]  # offset for left padding
+        text_to_overwrite = new_token_positions[batch_indices, non_image_indices]
+
+        # 3. Create the full embedding, already padded to the maximum position
+        final_embedding = torch.zeros(
+            batch_size, max_embed_dim, embed_dim, dtype=inputs_embeds.dtype, device=inputs_embeds.device
+        )
+        final_attention_mask = torch.zeros(
+            batch_size, max_embed_dim, dtype=attention_mask.dtype, device=inputs_embeds.device
+        )
+        if labels is not None:
+            final_labels = torch.full(
+                (batch_size, max_embed_dim), ignore_index, dtype=input_ids.dtype, device=input_ids.device
+            )
+        # In case the Vision model or the Language model has been offloaded to CPU, we need to manually
+        # set the corresponding tensors into their correct target device.
+        target_device = inputs_embeds.device
+        batch_indices, non_image_indices, text_to_overwrite = (
+            batch_indices.to(target_device),
+            non_image_indices.to(target_device),
+            text_to_overwrite.to(target_device),
+        )
+        if attention_mask is not None:
+            attention_mask = attention_mask.to(target_device)
+
+        # 4. Fill the embeddings based on the mask.
+        final_embedding[batch_indices, text_to_overwrite] = inputs_embeds[batch_indices, non_image_indices]
+        final_attention_mask[batch_indices, text_to_overwrite] = attention_mask[batch_indices, non_image_indices]
+        if labels is not None:
+            final_labels[batch_indices, text_to_overwrite] = labels[batch_indices, non_image_indices]
+
+        # 5. Fill the embeddings corresponding to the images. Anything that is not `text_positions` needs filling (#29835)
+        image_to_overwrite = torch.full(
+            (batch_size, max_embed_dim), True, dtype=torch.bool, device=inputs_embeds.device
+        )
+        image_to_overwrite[batch_indices, text_to_overwrite] = False
+        image_to_overwrite &= image_to_overwrite.cumsum(-1) - 1 >= nb_image_pad[:, None].to(target_device)
+
+        if image_to_overwrite.sum() != image_features.shape[:-1].numel():
+            raise ValueError(
+                f"The input provided to the model are wrong. The number of image tokens is {image_to_overwrite.sum()} while"
+                f" the number of image features given to the model is {image_features.shape[:-1].numel()}. "
+                "This prevents correct indexing and breaks batch generation."
+            )
+
+        final_embedding[image_to_overwrite] = image_features.contiguous().reshape(-1, embed_dim).to(target_device)
+        final_attention_mask |= image_to_overwrite
+        position_ids = (final_attention_mask.cumsum(-1) - 1).masked_fill_((final_attention_mask == 0), 1)
+
+        # 6. Mask out the embedding at padding positions, as we later use the past_key_value value to determine the non-attended tokens.
+        batch_indices, pad_indices = torch.where(input_ids == pad_token_id)
+        indices_to_mask = new_token_positions[batch_indices, pad_indices]
+
+        final_embedding[batch_indices, indices_to_mask] = 0
+
+        if labels is None:
+            final_labels = None
+        return final_embedding, final_attention_mask, final_labels, position_ids
+
+    def _extract_image_features(self,
+                                pixel_values: torch.FloatTensor | list[torch.FloatTensor],
+                                image_grid_thw: torch.FloatTensor,
+                                ):
+        """
+        Args:
+            pixel_values (:obj:`torch.FloatTensor` of shape :obj:`(sum_num_image_tokens, channels)`):
+                The pixel values of the images processed by image processor.
+            image_grid_thw: (B,3)
+
+        Returns:
+            selected_image_feature (:obj:`torch.FloatTensor` of shape :obj:`(num_image_tokens, embed_dim)`):
+                The selected image features to use as input to the projector head.
+            
+        """
+
+        assert len(image_grid_thw.shape)==2 and image_grid_thw.shape[1]==3, f"image_grid_thw must be a 2D tensor with shape (batched, 3), but got {image_grid_thw.shape}"
+        if isinstance(pixel_values, list):
+            pixel_values = torch.cat(pixel_values, dim=0)
+        image_features_ = self.vision_tower(pixel_values, grid_thw=image_grid_thw)
+        image_features_list = []
+        start_idx = 0
+        for i, grid_thw in enumerate(image_grid_thw):
+            end_idx = start_idx + (grid_thw[0] * grid_thw[1] * grid_thw[2]) // 4
+            image_features_list.append(image_features_[start_idx:end_idx, :])
+            start_idx = end_idx
+
+        selected_image_feature = torch.cat(image_features_list, dim=0)
+        feature_lengths = [x.size(0) for x in image_features_list]
+        return selected_image_feature, feature_lengths
+
+    def get_placeholder_mask(
+        self,
+        input_ids: torch.LongTensor,
+        inputs_embeds: torch.FloatTensor,
+        image_features: Optional[torch.FloatTensor] = None,
+        video_features: Optional[torch.FloatTensor] = None,
+    ):
+        """
+        Obtains multimodal placeholder mask from `input_ids` or `inputs_embeds`, and checks that the placeholder token count is
+        equal to the length of multimodal features. If the lengths are different, an error is raised.
+        """
+        if input_ids is None:
+            special_image_mask = inputs_embeds == self.get_input_embeddings()(
+                torch.tensor(self.config.media_placeholder_token_id, dtype=torch.long, device=inputs_embeds.device)
+            )
+            special_image_mask = special_image_mask.all(-1)
+        else:
+            special_image_mask = input_ids == self.config.media_placeholder_token_id
+
+        n_image_tokens = special_image_mask.sum()
+        special_image_mask = special_image_mask.unsqueeze(-1).expand_as(inputs_embeds).to(inputs_embeds.device)
+        if image_features is not None and inputs_embeds[special_image_mask].numel() != image_features.numel():
+            raise ValueError(
+                f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {image_features.shape[0]}"
+            )
+
+        return special_image_mask, None
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        pixel_values: torch.FloatTensor | list[torch.FloatTensor] | None = None,
+        image_grid_thw: torch.Tensor = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: list[torch.FloatTensor] | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        return_dict: bool | None = None,
+        **kwargs,
+    ) -> tuple | LlavaCausalLMOutputWithPast:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        ```"""
+        if attention_mask is None:
+            attention_mask = torch.ones_like(input_ids)
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if inputs_embeds is None:
+            # 1. Extra the input embeddings
+            inputs_embeds = self.get_input_embeddings()(input_ids)
+
+            # # 2. Merge text and images
+            # if pixel_values is not None and len(pixel_values) > 0 and input_ids.shape[1] != 1:
+            #     image_feature, feature_lengths = self._extract_image_features(
+            #         pixel_values, image_grid_thw)
+
+            #     inputs_embeds = inputs_embeds.to(image_feature.dtype)  # num_tokens, embed_dim
+            #     inputs_embeds, attention_mask, labels, position_ids = \
+            #         self._merge_input_ids_with_image_features(image_feature, feature_lengths, inputs_embeds, input_ids, attention_mask, labels
+            #         )
+
+            # FIXME: build image embeddings without merging
+            if pixel_values is not None and len(pixel_values) > 0 and input_ids.shape[1] != 1:
+                image_embeds, feature_lengths = self._extract_image_features(pixel_values, image_grid_thw)
+                image_mask, _ = self.get_placeholder_mask(
+                    input_ids, inputs_embeds=inputs_embeds, image_features=image_embeds
+                )
+                inputs_embeds = inputs_embeds.to(image_embeds.dtype)
+                inputs_embeds = inputs_embeds.masked_scatter(image_mask, image_embeds)
+
+            # In case input_ids.shape[1] == 1 & pixel_values==None & past_key_values != None, we are in the case of
+            # generation with cache
+            elif past_key_values is not None and pixel_values is not None and input_ids.shape[1] == 1:
+                # Retrieve the first layer to inspect the logits and mask out the hidden states
+                # that are set to 0
+                first_layer_past_key_value = past_key_values[0][0][:, :, :, 0]
+
+                # Sum all dimensions of head_dim (-2) to avoid random errors such as: https://github.com/huggingface/transformers/pull/28032#issuecomment-1863691941
+                batch_index, non_attended_tokens = torch.where(first_layer_past_key_value.float().sum(-2) == 0)
+
+                # Get the target length
+                target_length = input_ids.shape[1]
+                past_length = first_layer_past_key_value.shape[-1]
+
+                extended_attention_mask = torch.ones(
+                    (attention_mask.shape[0], past_length),
+                    dtype=attention_mask.dtype,
+                    device=attention_mask.device,
+                )
+
+                # Filter out only the tokens that can be un-attended, this can happen
+                # if one uses Llava + Fused modules where the cache on the
+                # first iteration is already big enough, or if one passes custom cache
+                valid_indices = non_attended_tokens < extended_attention_mask.size(-1)
+                new_batch_index = batch_index[valid_indices]
+                new_non_attended_tokens = non_attended_tokens[valid_indices]
+
+                # Zero-out the places where we don't need to attend
+                extended_attention_mask[new_batch_index, new_non_attended_tokens] = 0
+
+                attention_mask = torch.cat((extended_attention_mask, attention_mask[:, -target_length:]), dim=1)
+                position_ids = torch.sum(attention_mask, dim=1).unsqueeze(-1) - 1
+        cu_seqlens = kwargs.pop("cu_seqlens", None)
+        outputs = self.language_model(
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            cu_seqlens=cu_seqlens,
+            **kwargs,
+        )
+
+        logits = outputs[0]
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            if attention_mask is not None:
+                shift_attention_mask = attention_mask[..., 1:]
+                shift_logits = logits[..., :-1, :][shift_attention_mask.to(logits.device) != 0].contiguous()
+                shift_labels = labels[..., 1:][shift_attention_mask.to(labels.device) != 0].contiguous()
+            else:
+                shift_logits = logits[..., :-1, :].contiguous()
+                shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(
+                shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1).to(shift_logits.device)
+            )
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return LlavaCausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self, input_ids, past_key_values=None, inputs_embeds=None, pixel_values=None, image_grid_thw=None, attention_mask=None, **kwargs
+    ):
+        if past_key_values is not None:
+            if isinstance(past_key_values, Cache):
+                cache_length = past_key_values.get_seq_length()
+                past_length = past_key_values.seen_tokens if hasattr(past_key_values, 'seen_tokens') else past_key_values.get_seq_length()
+            else:
+                cache_length = past_length = past_key_values[0][0].shape[2]
+
+            # Keep only the unprocessed tokens:
+            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
+            # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
+            # input)
+            if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]:
+                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
+            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
+            # input_ids based on the past_length.
+            elif past_length < input_ids.shape[1]:
+                input_ids = input_ids[:, past_length:]
+            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.
+            elif self.config.media_placeholder_token_id in input_ids:
+                input_ids = input_ids[:, input_ids.shape[1] - 1 :]
+            # If the cache has seen more tokens than it can hold, then the cache has a size limit. Let's discard the
+            # older attention values, as their corresponding values are not part of the input.
+            if cache_length < past_length and attention_mask is not None:
+                attention_mask = attention_mask[:, -(cache_length + input_ids.shape[1]) :]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1] :]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+                "pixel_values": pixel_values,
+                "image_grid_thw": image_grid_thw,
+            }
+        )
+        return model_inputs
+
+    def _reorder_cache(self, *args, **kwargs):
+        return self.language_model._reorder_cache(*args, **kwargs)

preprocessor_config.json

@@ -0,0 +1,18 @@
+{
+  "min_pixels": 3136,
+  "max_pixels": 12845056,
+  "patch_size": 14,
+  "temporal_patch_size": 2,
+  "merge_size": 2,
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "image_processor_type": "Qwen2VLImageProcessor"
+}

processing_opencua.py

@@ -0,0 +1,270 @@
+from transformers.models.qwen2_5_vl.processing_qwen2_5_vl import Qwen2_5_VLProcessor
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache
+from transformers.configuration_utils import PretrainedConfig
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.image_utils import ImageInput
+from transformers.modeling_flash_attention_utils import is_flash_attn_available
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.processing_utils import ImagesKwargs, MultiModalData, ProcessingKwargs, ProcessorMixin, Unpack, VideosKwargs
+from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
+from transformers.utils import is_torchdynamo_compiling, logging
+from transformers.video_utils import VideoInput
+
+from transformers.models.qwen2_5_vl.processing_qwen2_5_vl import Qwen2_5_VLProcessorKwargs
+
+import torch
+import numpy as np
+from typing import Union, Optional
+
+# from typing import Union, Optional, TypedDict
+from PIL import Image
+
+if is_flash_attn_available():
+    pass
+
+logger = logging.get_logger(__name__)
+
+# class Qwen2_5_VLVideosProcessorKwargs(VideosKwargs, total=False):
+#     fps: Union[list[float], float]
+
+# class TokenizerChatTemplateKwargs(TypedDict, total=False):
+#     """
+#     Keyword arguments for tokenizer's `apply_chat_template`, when it is called from within a processor.
+
+#     tools (`list[Dict]`, *optional*):
+#         A list of tools (callable functions) that will be accessible to the model. If the template does not
+#         support function calling, this argument will have no effect. Each tool should be passed as a JSON Schema,
+#         giving the name, description and argument types for the tool. See our
+#         [chat templating guide](https://huggingface.co/docs/transformers/main/en/chat_templating#automated-function-conversion-for-tool-use)
+#         for more information.
+#     documents (`list[dict[str, str]]`, *optional*):
+#         A list of dicts representing documents that will be accessible to the model if it is performing RAG
+#         (retrieval-augmented generation). If the template does not support RAG, this argument will have no
+#         effect. We recommend that each document should be a dict containing "title" and "text" keys. Please
+#         see the RAG section of the [chat templating guide](https://huggingface.co/docs/transformers/main/en/chat_templating#arguments-for-RAG)
+#         for examples of passing documents with chat templates.
+#     add_generation_prompt (bool, *optional*):
+#         If this is set, a prompt with the token(s) that indicate
+#         the start of an assistant message will be appended to the formatted output. This is useful when you want to generate a response from the model.
+#         Note that this argument will be passed to the chat template, and so it must be supported in the
+#         template for this argument to have any effect.
+#     continue_final_message (bool, *optional*):
+#         If this is set, the chat will be formatted so that the final
+#         message in the chat is open-ended, without any EOS tokens. The model will continue this message
+#         rather than starting a new one. This allows you to "prefill" part of
+#         the model's response for it. Cannot be used at the same time as `add_generation_prompt`.
+#     return_assistant_tokens_mask (`bool`, defaults to `False`):
+#         Whether to return a mask of the assistant generated tokens. For tokens generated by the assistant,
+#         the mask will contain 1. For user and system tokens, the mask will contain 0.
+#         This functionality is only available for chat templates that support it via the `{% generation %}` keyword.
+#     """
+
+#     tools: Optional[list[dict]] = None
+#     documents: Optional[list[dict[str, str]]] = None
+#     add_generation_prompt: Optional[bool] = False
+#     continue_final_message: Optional[bool] = False
+#     return_assistant_tokens_mask: Optional[bool] = False
+
+# class ChatTemplateLoadKwargs(TypedDict, total=False):
+#     """
+#     Keyword arguments used to load multimodal data in processor chat templates.
+
+#     num_frames (`int`, *optional*):
+#         Number of frames to sample uniformly. If not passed, the whole video is loaded.
+#     load_audio_from_video (`bool`, *optional*):
+#             Whether to use the audio track of input video. If `True` the audio track will be loaded and passed to the
+#             processor. This flag has no effect if the model doesn't support audio modality.
+#     """
+
+#     sampling_rate: Optional[int] = 16_000
+#     load_audio_from_video: Optional[bool] = False
+
+
+# class ProcessorChatTemplateKwargs(ChatTemplateLoadKwargs, TokenizerChatTemplateKwargs, total=False):
+#     """
+#     Keyword arguments for processor's `apply_chat_template`.
+
+#     tokenize (`bool`, *optional*, defaults to `False`):
+#         Whether to tokenize the output or not.
+#     return_dict (`bool`, defaults to `False`):
+#         Whether to return a dictionary with named outputs. Has no effect if tokenize is `False`.
+#     """
+
+#     tokenize: Optional[bool] = False
+#     return_dict: Optional[bool] = False
+
+
+# class AllKwargsForChatTemplate(TypedDict, total=False):
+#     processor_kwargs: ProcessingKwargs
+#     mm_load_kwargs: ChatTemplateLoadKwargs
+#     template_kwargs: ProcessorChatTemplateKwargs
+
+# class Qwen2_5_VLImagesKwargs(ImagesKwargs):
+#     min_pixels: Optional[int]
+#     max_pixels: Optional[int]
+#     patch_size: Optional[int]
+#     temporal_patch_size: Optional[int]
+#     merge_size: Optional[int]
+
+
+# class Qwen2_5_VLProcessorKwargs(ProcessingKwargs, total=False):
+#     images_kwargs: Qwen2_5_VLImagesKwargs
+#     videos_kwargs: Qwen2_5_VLVideosProcessorKwargs
+#     _defaults = {
+#         "text_kwargs": {
+#             "padding": False,
+#             "return_mm_token_type_ids": False,
+#         },
+#     }
+
+class OpenCUAProcessor(Qwen2_5_VLProcessor):
+    attributes = ["image_processor", "tokenizer", "video_processor"]
+
+    image_processor_class = "AutoImageProcessor"
+    video_processor_class = "AutoVideoProcessor"
+    tokenizer_class = "AutoTokenizer"
+
+    def __init__(self,
+        image_processor: None,
+        tokenizer: None,
+        video_processor: None,
+        **kwargs,
+    ):
+        super().__init__(image_processor, tokenizer, video_processor, **kwargs)
+        self.image_token = "<|media_placeholder|>"  if not hasattr(tokenizer, "image_token") else tokenizer.image_token
+        self.video_token = "<|media_placeholder|>" if not hasattr(tokenizer, "video_token") else tokenizer.video_token
+
+        self.image_token_id = (
+            tokenizer.image_token_id
+            if getattr(tokenizer, "image_token_id", None)
+            else tokenizer.convert_tokens_to_ids(self.image_token)
+        )
+        self.video_token_id = (
+            tokenizer.video_token_id
+            if getattr(tokenizer, "video_token_id", None)
+            else tokenizer.convert_tokens_to_ids(self.video_token)
+        )
+        self.chat_template = self.tokenizer.chat_template
+        self.bos_token = self.tokenizer.bos_token
+        self.eos_token = self.tokenizer.eos_token
+        self.pad_token = self.tokenizer.pad_token
+        self.unk_token = self.tokenizer.unk_token
+    
+    def __call__(
+        self,
+        images: Optional[ImageInput] = None,
+        text: Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]] = None,
+        videos: Optional[VideoInput] = None,
+        **kwargs: Unpack[Qwen2_5_VLProcessorKwargs],
+    ) -> BatchFeature:
+        """
+        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
+        and `kwargs` arguments to Qwen2TokenizerFast's [`~Qwen2TokenizerFast.__call__`] if `text` is not `None` to encode
+        the text. To prepare the vision inputs, this method forwards the `vision_infos` and `kwargs` arguments to
+        Qwen2VLImageProcessor's [`~Qwen2VLImageProcessor.__call__`] if `vision_infos` is not `None`.
+
+        Args:
+            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `list[PIL.Image.Image]`, `list[np.ndarray]`, `list[torch.Tensor]`):
+                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
+                tensor. Both channels-first and channels-last formats are supported.
+            text (`str`, `list[str]`, `list[list[str]]`):
+                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
+                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
+                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
+            videos (`np.ndarray`, `torch.Tensor`, `list[np.ndarray]`, `list[torch.Tensor]`):
+                The image or batch of videos to be prepared. Each video can be a 4D NumPy array or PyTorch
+                tensor, or a nested list of 3D frames. Both channels-first and channels-last formats are supported.
+            return_tensors (`str` or [`~utils.TensorType`], *optional*):
+                If set, will return tensors of a particular framework. Acceptable values are:
+                - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                - `'np'`: Return NumPy `np.ndarray` objects.
+
+        Returns:
+            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
+
+            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
+            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
+              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
+              `None`).
+            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
+            - **pixel_values_videos** -- Pixel values of videos to be fed to a model. Returned when `videos` is not `None`.
+            - **image_grid_thw** -- List of image 3D grid in LLM. Returned when `images` is not `None`.
+            - **video_grid_thw** -- List of video 3D grid in LLM. Returned when `videos` is not `None`.
+            - **second_per_grid_ts** -- List of video seconds per time grid. Returned when `videos` is not `None`.
+        """
+        output_kwargs = self._merge_kwargs(
+            Qwen2_5_VLProcessorKwargs,
+            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
+            **kwargs,
+        )
+
+        image_inputs = videos_inputs = {}
+        if images is not None:
+            image_inputs = self.image_processor(images=images, **output_kwargs["images_kwargs"])
+            image_grid_thw = image_inputs["image_grid_thw"]
+
+        if videos is not None:
+            fps = output_kwargs["videos_kwargs"].get("fps", 2.0)
+            videos_inputs = self.video_processor(videos=videos, **output_kwargs["videos_kwargs"])
+            video_grid_thw = videos_inputs["video_grid_thw"]
+
+            if isinstance(fps, (int, float)):
+                second_per_grid_ts = [self.video_processor.temporal_patch_size / fps] * len(video_grid_thw)
+            elif hasattr(fps, "__len__") and len(fps) == len(video_grid_thw):
+                second_per_grid_ts = [self.video_processor.temporal_patch_size / tmp for tmp in fps]
+            else:
+                raise ValueError(
+                    f"The length of fps ({len(fps) if hasattr(fps, '__len__') else fps}) must be equal to the length of video_grid_thw ({len(video_grid_thw)}) or fps should be a single number."
+                )
+            videos_inputs.update({"second_per_grid_ts": second_per_grid_ts})
+
+        if not isinstance(text, list):
+            text = [text]
+            
+        text = text.copy()  # below lines change text in-place
+        if images is not None:
+            merge_length = self.image_processor.merge_size**2
+            index = 0
+            for i in range(len(text)):
+                while self.image_token in text[i]:
+                    num_image_tokens = image_grid_thw[index].prod() // merge_length
+                    text[i] = text[i].replace(self.image_token, '<|temp_placeholder|>' * num_image_tokens, 1)
+                    index += 1
+                text[i] = text[i].replace('<|temp_placeholder|>', self.image_token)
+
+        if videos is not None:
+            merge_length = self.video_processor.merge_size**2
+            index = 0
+            for i in range(len(text)):
+                while self.video_token in text[i]:
+                    num_video_tokens = video_grid_thw[index].prod() // merge_length
+                    text[i] = text[i].replace(self.video_token, '<|temp_placeholder|>' * num_video_tokens, 1)
+                    index += 1
+                text[i] = text[i].replace('<|temp_placeholder|>', self.video_token)
+
+        return_tensors = output_kwargs["text_kwargs"].pop("return_tensors", None)
+        return_mm_token_type_ids = output_kwargs["text_kwargs"].pop("return_mm_token_type_ids", None)
+        # from IPython import embed; embed()
+        text_inputs = self.tokenizer(text, **output_kwargs["text_kwargs"])
+        self._check_special_mm_tokens(text, text_inputs, modalities=["image", "video"])
+
+        if return_mm_token_type_ids:
+            array_ids = np.array(text_inputs["input_ids"])
+            mm_token_type_ids = np.zeros_like(text_inputs["input_ids"])
+            mm_token_type_ids[array_ids == self.image_token_id] = 1
+            text_inputs["mm_token_type_ids"] = mm_token_type_ids.tolist()
+
+        return BatchFeature(data={**text_inputs, **image_inputs, **videos_inputs}, tensor_type=return_tensors)
+
+
+#     @property
+#     def model_input_names(self):
+#         tokenizer_input_names = self.tokenizer.model_input_names
+#         image_processor_input_names = self.image_processor.model_input_names
+#         names_from_processor = list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
+#         return names_from_processor + ["second_per_grid_ts"]
+
+
+__all__ = ["OpenCUAProcessor"]

tiktoken.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2b1b8dfb5cc5f024bafc373121c6aba3f66f9a5a0269e243470a1de16a33186
+size 2561218

tokenization_opencua.py

@@ -0,0 +1,367 @@
+import os
+import tiktoken
+
+from logging import getLogger
+from pathlib import Path
+from typing import (
+    cast,
+    Tuple,
+    Dict,
+    Iterator,
+    List,
+    Union,
+    Optional,
+)
+from shutil import copyfile
+from tiktoken.load import load_tiktoken_bpe
+from tokenizers import AddedToken
+from transformers.tokenization_utils import PreTrainedTokenizer
+from transformers.models.gpt2.tokenization_gpt2 import bytes_to_unicode
+
+
+
+logger = getLogger(__name__)
+VOCAB_FILES_NAMES = {"vocab_file": "tiktoken.model"}
+
+class TikTokenTokenizer(PreTrainedTokenizer):
+    """
+    Tokenizing and encoding/decoding text using the Tiktoken tokenizer. See megatron/tokenizer/tiktoken_tokenizer.py.
+
+    This tokenizer inherits from [`PreTrainedTokenizer`] which contains most of the main methods. Users should refer to
+    this superclass for more information regarding those methods.
+
+    Args:
+        vocab_file (`str`):
+            The path to the Tiktoken model file.
+        bos_token (`str` or `tokenizers.AddedToken`, *optional*, defaults to `"<|begin_of_text|>",`):
+            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.
+        eos_token (`str` or `tokenizers.AddedToken`, *optional*, defaults to `"<|end_of_text|>"`):
+            The end of sequence token.
+        unk_token (`str` or `tokenizers.AddedToken`, *optional*, defaults to `"<|reserved_special_token_249|>"`):
+            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
+            token instead. The second to last item in special_tokens.
+        pad_token (`str` or `tokenizers.AddedToken`, *optional*, defaults to `"<|reserved_special_token_250|>"`):
+            The token used for padding, for example when batching sequences of different lengths.
+        additional_special_tokens (list of `str`, *optional*):
+            A tuple or a list of additional tokens, which will be marked as `special`, meaning that they will be
+            skipped when decoding if `skip_special_tokens` is set to `True`.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+
+    model_input_names = ["input_ids", "attention_mask"]
+
+    special_tokens: Dict[str, int]
+
+    num_reserved_special_tokens = 256
+
+    pat_str = "|".join(
+        [
+            r"""[\p{Han}]+""",
+            r"""[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}&&[^\p{Han}]]*[\p{Ll}\p{Lm}\p{Lo}\p{M}&&[^\p{Han}]]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?""",
+            r"""[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}&&[^\p{Han}]]+[\p{Ll}\p{Lm}\p{Lo}\p{M}&&[^\p{Han}]]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?""",
+            r"""\p{N}{1,3}""",
+            r""" ?[^\s\p{L}\p{N}]+[\r\n]*""",
+            r"""\s*[\r\n]+""",
+            r"""\s+(?!\S)""",
+            r"""\s+""",
+        ]
+    )
+
+    def __init__(
+        self,
+        vocab_file,
+        bos_token: Union[str, AddedToken]="[BOS]",
+        eos_token: Union[str, AddedToken]="[EOS]",
+        unk_token: Union[str, AddedToken, None]=None,
+        pad_token: Union[str, AddedToken, None]=None,
+        additional_special_tokens: List[str]=None,
+        added_tokens_decoder: Optional[dict] = None,
+        **kwargs,
+    ):
+        assert os.path.isfile(vocab_file), vocab_file
+
+        if additional_special_tokens is None:
+            # dumping mode
+            used_special_tokens  = [
+                "<|im_end|>", 
+                "<|im_user|>", 
+                "<|im_assistant|>", 
+                "<|reserved_token_0|>",
+                "<|start_header_id|>", 
+                "<|end_header_id|>", 
+                "<|reserved_token_1|>", 
+                "[EOT]",  
+                "<|im_system|>", 
+                "<|reserved_token_2|>",
+                "<|reserved_token_3|>",
+                "<|reserved_token_4|>",
+                "<|reserved_token_5|>",
+                "<|reserved_token_6|>", 
+                "<|reserved_token_7|>",
+                "<|im_middle|>", 
+                "<|media_begin|>",
+                "<|media_content|>", 
+                "<|media_end|>", 
+                "<|media_placeholder|>", 
+            ] 
+            used_reserved_tokens = 8
+            last_reserved_token_id = self.num_reserved_special_tokens - 4 - len(used_special_tokens)  + used_reserved_tokens - 1
+            additional_special_tokens = used_special_tokens + [
+                f"<|reserved_token_{i}|>"
+                for i in range(used_reserved_tokens, last_reserved_token_id + 1)
+            ]
+            # num_reserved_special_tokens = additional_special_tokens + BOS + EOS + unk_token + pad_token
+            assert len(additional_special_tokens) + 4 == self.num_reserved_special_tokens, f"additional_special_tokens num: {len(additional_special_tokens)} is not correct"
+            # we assume that the instance is under initialization and unk_token and pad_token should be automatically inferred
+            if unk_token is not None:
+                raise ValueError("unk_token should not be set in dumping mode when additional_special_tokens is None")
+            if pad_token is not None:
+                raise ValueError("pad_token should not be set in dumping mode when additional_special_tokens is None")
+            # last two reserved tokens
+            unk_token = f"[UNK]"
+            pad_token = f"[PAD]"
+
+            logger.info(f"adding unk_token: {unk_token} and pad_token: {pad_token}")
+            self.additional_special_tokens = additional_special_tokens
+            special_tokens = [str(bos_token), str(eos_token)] + additional_special_tokens + [str(unk_token), str(pad_token)]
+
+            self.vocab_file = vocab_file
+            mergeable_ranks = load_tiktoken_bpe(vocab_file)
+            num_base_tokens = len(mergeable_ranks)
+            self.special_tokens = {
+                token: num_base_tokens + i for i, token in enumerate(special_tokens)
+            }
+        else:
+            self.additional_special_tokens = additional_special_tokens
+            special_tokens_mapping = {
+            i: added_tokens_decoder[i].content for i in added_tokens_decoder
+        }
+
+            self.vocab_file = vocab_file
+            mergeable_ranks = load_tiktoken_bpe(vocab_file)
+            num_base_tokens = len(mergeable_ranks)
+            self.special_tokens = {
+                special_tokens_mapping.get(i, f"<|reserved_token_{i}|>"): i
+                for i in range(
+                    num_base_tokens, num_base_tokens + self.num_reserved_special_tokens + 2
+                )
+            }
+
+
+
+        self.model = tiktoken.Encoding(
+            name=Path(vocab_file).name,
+            pat_str=self.pat_str,
+            mergeable_ranks=mergeable_ranks,
+            special_tokens=self.special_tokens,
+        )
+        logger.info(f"Reloaded tiktoken model from {vocab_file}")
+
+        self.n_words: int = self.model.n_vocab
+        # BOS / EOS token IDs
+        self.bos_id: int = self.special_tokens[str(bos_token)]
+        self.eos_id: int = self.special_tokens[str(eos_token)]
+        
+        logger.info(
+            f"#words: {self.n_words} - BOS ID: {self.bos_id} - EOS ID: {self.eos_id}"
+        )
+
+        self.pad_id: int = self.special_tokens[str(pad_token)]
+        self.unk_id: int = self.special_tokens[str(unk_token)]
+        self.byte_encoder = bytes_to_unicode()
+        self.byte_decoder = {v: k for k, v in self.byte_encoder.items()}
+
+        self.decoder = {}
+        for i in range(self.n_words):
+            # Taken from https://gist.github.com/xenova/a452a6474428de0182b17605a98631ee
+            decoding = ''.join([
+                self.byte_encoder[ord(char)] for char in
+                self.model.decode_single_token_bytes(i).decode('latin-1')
+            ])
+            self.decoder[i] = decoding
+
+        self.encoder = {}
+        for i in range(self.n_words):
+            if i in self.decoder:
+                self.encoder[self.decoder[i]] = i
+
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            pad_token=pad_token,
+            additional_special_tokens=self.additional_special_tokens,
+            **kwargs,
+        )
+        self.all_special_ids_set = set(self.all_special_ids)
+
+    def encode(
+        self,
+        text: str,
+        allow_special_tokens = True,
+        **kwargs
+    ) -> List[int]:
+        """
+        Encodes a string into a list of token IDs.
+
+        Args:
+            text (str): The input string to be encoded.
+
+        Returns:
+            list[int]: A list of token IDs.
+        """
+        # If there are other args, we should call super().encode because there are a lot of code
+        # to handle those args. supper().encode finally will call _tokenize and _convert_token_to_id.
+        # NOTE: our encode method is not compatible with the super().encode method,
+        #   e.g. split_special_tokens' default is True in our encode method.
+        if len(kwargs) > 0:
+            logger.warning( f"Calling super().encode with {kwargs}" )
+            return super().encode(text, **kwargs)
+
+        assert type(text) is str
+
+        # The tiktoken tokenizer can handle <=400k chars without
+        # pyo3_runtime.PanicException.
+        TIKTOKEN_MAX_ENCODE_CHARS = 400_000
+
+        # https://github.com/openai/tiktoken/issues/195
+        # Here we iterate over subsequences and split if we exceed the limit
+        # of max consecutive non-whitespace or whitespace characters.
+        MAX_NO_WHITESPACES_CHARS = 25_000
+
+        texts = self.pre_tokenizer_process(text)
+
+        all_substrs = []
+        for text in texts:
+            substrs = (
+                substr
+                for i in range(0, len(text), TIKTOKEN_MAX_ENCODE_CHARS)
+                for substr in self._split_whitespaces_or_nonwhitespaces(
+                    text[i: i + TIKTOKEN_MAX_ENCODE_CHARS], MAX_NO_WHITESPACES_CHARS
+                )
+            )
+            all_substrs.extend(substrs)
+
+        t: List[int] = []
+        for substr in all_substrs:
+            if allow_special_tokens:
+                t.extend(
+                    self.model.encode(
+                        substr,
+                        allowed_special="all",
+                    )
+                )
+            else:
+                t.extend(
+                    self.model.encode(
+                        substr,
+                        disallowed_special=(),
+                    )
+                )
+
+        return t
+
+    def decode(
+        self,
+        token_ids: Union[int, List[int]],
+        **kwargs
+    ) -> str:
+        """
+        Decodes a list of token IDs into a string.
+
+        Args:
+            token_ids (List[int]): The list of token IDs to be decoded.
+
+        Returns:
+            str: The decoded string.
+        """
+        # If there are other args, we should call super().decode because there are a lot of code
+        # to handle those args. supper().encode finally will call convert_tokens_to_string and _convert_id_to_token.
+        if len(kwargs) > 0:
+            return super().decode(token_ids, **kwargs)
+
+        if type(token_ids) is int:
+            token_ids = [token_ids]
+
+        return self.model.decode(cast(List[int], token_ids))
+
+    @staticmethod
+    def _split_whitespaces_or_nonwhitespaces(
+        s: str, max_consecutive_slice_len: int
+    ) -> Iterator[str]:
+        """
+        Splits the string `s` so that each substring contains no more than `max_consecutive_slice_len`
+        consecutive whitespaces or consecutive non-whitespaces.
+        """
+        current_slice_len = 0
+        current_slice_is_space = s[0].isspace() if len(s) > 0 else False
+        slice_start = 0
+
+        for i in range(len(s)):
+            is_now_space = s[i].isspace()
+
+            if current_slice_is_space ^ is_now_space:
+                current_slice_len = 1
+                current_slice_is_space = is_now_space
+            else:
+                current_slice_len += 1
+                if current_slice_len > max_consecutive_slice_len:
+                    yield s[slice_start:i]
+                    slice_start = i
+                    current_slice_len = 1
+        yield s[slice_start:]
+
+    def pre_tokenizer_process(self, text: str) -> List[str]:
+        """
+        pre-tokenizes the input text into a list of tokens.
+        This method is used to split the input text into smaller chunks for internal processing.
+        """
+        return [text]
+
+
+    """ ----- Below are the abstract methods required by PreTrainedTokenizer ----- """
+    @property
+    def vocab_size(self) -> int:
+        return self.n_words
+
+    def get_vocab(self) -> Dict[str, int]:
+        return self.encoder
+
+    def _tokenize(self, text: str, **kwargs) -> List[str]:
+        return [
+            self.decoder[t]
+            for t in self.encode(text)
+        ]
+
+    def _convert_token_to_id(self, token: str) -> int:
+        return self.encoder.get(token, self.unk_id)
+
+    def _convert_id_to_token(self, index: int) -> str:
+        return self.decoder.get(index)
+
+    @staticmethod
+    def clean_up_tokenization(out_string: str) -> str:
+        return out_string
+
+    def convert_tokens_to_string(self, tokens: List[str]) -> str:
+        text = ''.join(tokens)
+        text = bytearray([self.byte_decoder[c] for c in text]).decode('utf-8', 'replace')
+        return text
+
+    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        if not os.path.isdir(save_directory):
+            raise ValueError(f"vocabulary path ({save_directory}) should be a directory")
+        out_vocab_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
+        )
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+
+        return (out_vocab_file,)
+
+
+class TikTokenV3(TikTokenTokenizer):
+    num_reserved_special_tokens = 293 + 128
+    pat_str = "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+"

tokenizer_config.json

@@ -0,0 +1,234 @@
+{
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "[BOS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "[EOS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|im_user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|im_assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|reserved_token_0|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|start_header_id|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|end_header_id|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|reserved_token_1|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "[EOT]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|im_system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|reserved_token_2|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|reserved_token_3|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|reserved_token_4|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<|reserved_token_5|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151658": {
+      "content": "<|reserved_token_6|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151659": {
+      "content": "<|reserved_token_7|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151660": {
+      "content": "<|im_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151661": {
+      "content": "<|media_begin|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151662": {
+      "content": "<|media_content|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151663": {
+      "content": "<|media_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151664": {
+      "content": "<|media_placeholder|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "152062": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "152063": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+
+  },
+  "additional_special_tokens": [
+    "<|im_end|>",
+    "<|im_user|>", 
+    "<|im_assistant|>", 
+    "<|reserved_token_0|>",
+    "<|start_header_id|>", 
+    "<|end_header_id|>", 
+    "<|reserved_token_1|>", 
+    "[EOT]",  
+    "<|im_system|>", 
+    "<|reserved_token_2|>", 
+    "<|reserved_token_3|>", 
+    "<|reserved_token_4|>", 
+    "<|reserved_token_5|>",
+    "<|reserved_token_6|>",
+    "<|reserved_token_7|>",
+    "<|im_middle|>", 
+    "<|media_begin|>", 
+    "<|media_content|>", 
+    "<|media_end|>", 
+    "<|media_placeholder|>"
+  ],
+  "bos_token": "[BOS]",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "[EOS]",
+  "extra_special_tokens": {},
+  "chat_template": "{%- for message in messages -%}{%- if loop.first and messages[0]['role'] != 'system' -%}{{'<|im_system|>system<|im_middle|>You are a helpful assistant<|im_end|>'}}{%- endif -%}{%- if message['role'] == 'system' -%}{{'<|im_system|>'}}{%- endif -%}{%- if message['role'] == 'user' -%}{{'<|im_user|>'}}{%- endif -%}{%- if message['role'] == 'assistant' -%}{{'<|im_assistant|>'}}{%- endif -%}{{- message['role'] -}}{{'<|im_middle|>'}}{%- if message['content'] is string -%}{{- message['content'] + '<|im_end|>' -}}{%- else -%}{%- for content in message['content'] -%}{%- if content['type'] == 'image' or 'image' in content or 'image_url' in content -%}{{'<|media_begin|>image<|media_content|><|media_placeholder|><|media_end|>'}}{%- else -%}{{content['text']}}{%- endif -%}{%- endfor -%}{{'<|im_end|>'}}{%- endif -%}{%- endfor -%}{%- if add_generation_prompt -%}{{'<|im_assistant|>assistant<|im_middle|>'}}{%- endif -%}",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "[PAD]",
+  "tokenizer_class": "TikTokenV3",
+  "unk_token": "[UNK]",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_opencua.TikTokenV3",
+      null
+    ]
+  }
+}