Upload 15 files

QUANTISED FP16 BNB version (aprox 8Gb)

args.json

@@ -0,0 +1,28 @@
+{
+    "path_to_vm": "/home/sujit/VirtualBox VMs/ubuntu24.04/ubuntu24.04.vbox",
+    "headless": true,
+    "action_space": "pyautogui",
+    "observation_type": "screenshot",
+    "sleep_after_execution": 3.0,
+    "max_steps": 50,
+    "test_config_base_dir": "evaluation_examples",
+    "model": "/home/sujit/OpenCUA-7B",
+    "temperature": 0,
+    "top_p": 0.9,
+    "max_tokens": 2048,
+    "stop_token": null,
+    "cot_level": "l2",
+    "history_type": "action_history",
+    "coordinate_type": "qwen25",
+    "max_image_history_length": 3,
+    "domain": "all",
+    "test_all_meta_path": "evaluation_examples/test_small.json",
+    "result_dir": "./results",
+    "num_envs": 1,
+    "log_level": "INFO",
+    "region": "us-east-1",
+    "provider_name": "virtualbox",
+    "client_password": "password",
+    "screen_width": 1920,
+    "screen_height": 1080
+}

bnb_snapshot.json

@@ -0,0 +1,3 @@
+{
+  "already_quantized": true
+}

config.json

@@ -0,0 +1,260 @@
+{
+  "vision_config": {
+    "return_dict": true,
+    "output_hidden_states": false,
+    "output_attentions": false,
+    "torchscript": false,
+    "torch_dtype": null,
+    "use_bfloat16": false,
+    "tf_legacy_loss": false,
+    "pruned_heads": {},
+    "tie_word_embeddings": true,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "is_decoder": false,
+    "cross_attention_hidden_size": null,
+    "add_cross_attention": false,
+    "tie_encoder_decoder": false,
+    "max_length": 20,
+    "min_length": 0,
+    "do_sample": false,
+    "early_stopping": false,
+    "num_beams": 1,
+    "num_beam_groups": 1,
+    "diversity_penalty": 0.0,
+    "temperature": 1.0,
+    "top_k": 50,
+    "top_p": 1.0,
+    "typical_p": 1.0,
+    "repetition_penalty": 1.0,
+    "length_penalty": 1.0,
+    "no_repeat_ngram_size": 0,
+    "encoder_no_repeat_ngram_size": 0,
+    "bad_words_ids": null,
+    "num_return_sequences": 1,
+    "output_scores": false,
+    "return_dict_in_generate": false,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "remove_invalid_values": false,
+    "exponential_decay_length_penalty": null,
+    "suppress_tokens": null,
+    "begin_suppress_tokens": null,
+    "architectures": null,
+    "finetuning_task": null,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "tokenizer_class": null,
+    "prefix": null,
+    "bos_token_id": null,
+    "pad_token_id": null,
+    "eos_token_id": null,
+    "sep_token_id": null,
+    "decoder_start_token_id": null,
+    "task_specific_params": null,
+    "problem_type": null,
+    "_name_or_path": "",
+    "_attn_implementation_autoset": false,
+    "in_chans": 3,
+    "spatial_patch_size": 14,
+    "depth": 32,
+    "hidden_size": 1280,
+    "hidden_act": "silu",
+    "intermediate_size": 3420,
+    "num_heads": 16,
+    "in_channels": 3,
+    "patch_size": 14,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2,
+    "tokens_per_second": 2,
+    "window_size": 112,
+    "fullatt_block_indexes": [
+      7,
+      15,
+      23,
+      31
+    ],
+    "out_hidden_size": 3584,
+    "model_type": "qwen2_5_vl"
+  },
+  "text_config": {
+    "vocab_size": 152064,
+    "max_position_embeddings": 32768,
+    "hidden_size": 3584,
+    "intermediate_size": 18944,
+    "num_hidden_layers": 28,
+    "num_attention_heads": 28,
+    "use_sliding_window": false,
+    "sliding_window": 4096,
+    "max_window_layers": 28,
+    "num_key_value_heads": 4,
+    "hidden_act": "silu",
+    "initializer_range": 0.02,
+    "rms_norm_eps": 1e-05,
+    "use_cache": true,
+    "rope_theta": 1000000.0,
+    "rope_scaling": null,
+    "attention_dropout": 0.0,
+    "return_dict": true,
+    "output_hidden_states": false,
+    "output_attentions": false,
+    "torchscript": false,
+    "torch_dtype": "bfloat16",
+    "use_bfloat16": false,
+    "tf_legacy_loss": false,
+    "pruned_heads": {},
+    "tie_word_embeddings": false,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "is_decoder": false,
+    "cross_attention_hidden_size": null,
+    "add_cross_attention": false,
+    "tie_encoder_decoder": false,
+    "max_length": 20,
+    "min_length": 0,
+    "do_sample": false,
+    "early_stopping": false,
+    "num_beams": 1,
+    "num_beam_groups": 1,
+    "diversity_penalty": 0.0,
+    "temperature": 1.0,
+    "top_k": 50,
+    "top_p": 1.0,
+    "typical_p": 1.0,
+    "repetition_penalty": 1.0,
+    "length_penalty": 1.0,
+    "no_repeat_ngram_size": 0,
+    "encoder_no_repeat_ngram_size": 0,
+    "bad_words_ids": null,
+    "num_return_sequences": 1,
+    "output_scores": false,
+    "return_dict_in_generate": false,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "remove_invalid_values": false,
+    "exponential_decay_length_penalty": null,
+    "suppress_tokens": null,
+    "begin_suppress_tokens": null,
+    "architectures": null,
+    "finetuning_task": null,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "tokenizer_class": null,
+    "prefix": null,
+    "bos_token_id": 151643,
+    "pad_token_id": 152063,
+    "eos_token_id": 151644,
+    "sep_token_id": null,
+    "decoder_start_token_id": null,
+    "task_specific_params": null,
+    "problem_type": null,
+    "_name_or_path": "",
+    "_attn_implementation_autoset": false,
+    "head_dim": 128,
+    "k_proj_bias": true,
+    "model_type": "qwen2",
+    "pretraining_sequence_length": 128000,
+    "q_proj_bias": true,
+    "v_proj_bias": true
+  },
+  "ignore_index": -100,
+  "media_placeholder_token_id": 151664,
+  "return_dict": true,
+  "output_hidden_states": false,
+  "output_attentions": false,
+  "torchscript": false,
+  "torch_dtype": "bfloat16",
+  "use_bfloat16": false,
+  "tf_legacy_loss": false,
+  "pruned_heads": {},
+  "tie_word_embeddings": false,
+  "chunk_size_feed_forward": 0,
+  "is_encoder_decoder": false,
+  "is_decoder": false,
+  "cross_attention_hidden_size": null,
+  "add_cross_attention": false,
+  "tie_encoder_decoder": false,
+  "max_length": 20,
+  "min_length": 0,
+  "do_sample": false,
+  "early_stopping": false,
+  "num_beams": 1,
+  "num_beam_groups": 1,
+  "diversity_penalty": 0.0,
+  "temperature": 1.0,
+  "top_k": 50,
+  "top_p": 1.0,
+  "typical_p": 1.0,
+  "repetition_penalty": 1.0,
+  "length_penalty": 1.0,
+  "no_repeat_ngram_size": 0,
+  "encoder_no_repeat_ngram_size": 0,
+  "bad_words_ids": null,
+  "num_return_sequences": 1,
+  "output_scores": false,
+  "return_dict_in_generate": false,
+  "forced_bos_token_id": null,
+  "forced_eos_token_id": null,
+  "remove_invalid_values": false,
+  "exponential_decay_length_penalty": null,
+  "suppress_tokens": null,
+  "begin_suppress_tokens": null,
+  "architectures": [
+    "OpenCUAForConditionalGeneration"
+  ],
+  "finetuning_task": null,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1"
+  },
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1
+  },
+  "tokenizer_class": null,
+  "prefix": null,
+  "bos_token_id": null,
+  "pad_token_id": 0,
+  "eos_token_id": null,
+  "sep_token_id": null,
+  "decoder_start_token_id": null,
+  "task_specific_params": null,
+  "problem_type": null,
+  "_name_or_path": "/home/sujit/OpenCUA-7B",
+  "_attn_implementation_autoset": false,
+  "transformers_version": "4.49.0",
+  "auto_map": {
+    "AutoConfig": "configuration_opencua.OpenCUAConfig",
+    "AutoModel": "modeling_opencua.OpenCUAForConditionalGeneration",
+    "AutoModelForCausalLM": "modeling_opencua.OpenCUAForConditionalGeneration"
+  },
+  "model_type": "opencua",
+  "vocab_size": 152064,
+  "quantization_config": {
+    "_load_in_4bit": true,
+    "_load_in_8bit": false,
+    "bnb_4bit_compute_dtype": "float16",
+    "bnb_4bit_quant_storage": "uint8",
+    "bnb_4bit_quant_type": "nf4",
+    "bnb_4bit_use_double_quant": true,
+    "llm_int8_enable_fp32_cpu_offload": false,
+    "llm_int8_has_fp16_weight": false,
+    "llm_int8_skip_modules": null,
+    "llm_int8_threshold": 6.0,
+    "load_in_4bit": true,
+    "load_in_8bit": false,
+    "quant_method": "bitsandbytes"
+  }
+}

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 +1,7 @@
+{
+  "_attn_implementation": "sdpa",
+  "attn_implementation": "sdpa",
+  "eos_token_id": 151644,
+  "max_length": 32768,
+  "transformers_version": "4.49.0"
+}

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1bcbdd077a2e97bf65a6beaff1fee1327aec3665b6a5b8fc68131bfe2ab9cb0
+size 4809638813

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9be0baa3e20f5a2bcf0f0dee07c9586336a48fcacb297944fc5e449bec9ba3d
+size 1089994896

modeling_opencua.py

@@ -0,0 +1,449 @@
+# ------------------------------------------------------------------------------
+# 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
+
+
+class OpenCUAPreTrainedModel(PreTrainedModel):
+    config_class = OpenCUAConfig
+    base_model_prefix = "model"
+    _no_split_modules = ["Qwen2_5_VisionTransformerPretrainedModel"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = 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_()
+
+    @property
+    def _supports_sdpa(self):
+        """
+        Retrieve language_model's attribute to check whether the model supports
+        SDPA or not.
+        """
+        return self.language_model._supports_sdpa
+
+
+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.post_init()
+
+    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),
+        )
+        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],
+                                grid_thws: 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.
+            grid_thws: (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(grid_thws.shape)==2 and grid_thws.shape[1]==3, f"grid_thws must be a 2D tensor with shape (batched, 3), but got {grid_thws.shape}"
+        if isinstance(pixel_values, list):
+            pixel_values = torch.cat(pixel_values, dim=0)
+        image_features_ = self.vision_tower(pixel_values, grid_thw=grid_thws)
+        image_features_list = []
+        start_idx = 0
+        for i, grid_thw in enumerate(grid_thws):
+            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 forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        pixel_values: torch.FloatTensor | list[torch.FloatTensor] | None = None,
+        grid_thws: 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,
+    ) -> 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]`.
+
+        ```"""
+
+        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, grid_thws)
+
+                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
+                    )
+            # 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
+
+        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,
+        )
+
+        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, grid_thws=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
+            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,
+                "grid_thws": grid_thws,
+            }
+        )
+        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"
+}

quantization_config.json

@@ -0,0 +1,16 @@
+{
+  "quant_method": "bitsandbytes",
+  "_load_in_8bit": false,
+  "_load_in_4bit": true,
+  "llm_int8_threshold": 6.0,
+  "llm_int8_skip_modules": null,
+  "llm_int8_enable_fp32_cpu_offload": false,
+  "llm_int8_has_fp16_weight": false,
+  "bnb_4bit_quant_type": "nf4",
+  "bnb_4bit_use_double_quant": true,
+  "bnb_4bit_compute_dtype": "fp16",
+  "bnb_4bit_quant_storage": "uint8",
+  "load_in_4bit": true,
+  "load_in_8bit": false,
+  "is_pre_quantized": true
+}

special_tokens_map.json

@@ -0,0 +1,52 @@
+{
+  "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": {
+    "content": "[BOS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "[EOS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

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
+    ]
+  }
+}