# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # This file was automatically generated from src/transformers/models/qwen3_vl/modular_qwen3_vl.py. # Do NOT edit this file manually as any edits will be overwritten by the generation of # the file from the modular. If any change should be done, please apply the change to the # modular_qwen3_vl.py file directly. One of our CI enforces this. # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # coding=utf-8 # Copyright 2025 The Qwen Team and The HuggingFace Inc. team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Union,Optional import numpy as np from transformers.feature_extraction_utils import BatchFeature from transformers.image_utils import ImageInput from transformers.processing_utils import MultiModalData, ProcessingKwargs, ProcessorMixin, Unpack from transformers.tokenization_utils_base import PreTokenizedInput, TextInput from transformers.utils import logging from transformers.video_utils import VideoInput import os logger = logging.get_logger(__name__) class InternS1ProProcessorKwargs(ProcessingKwargs, total=False): _defaults = { "text_kwargs": { "padding": False, "return_token_type_ids": False, "return_mm_token_type_ids": False, }, "videos_kwargs": {"return_metadata": True}, "time_series_kwargs": {}, } class InternS1ProProcessor(ProcessorMixin): r""" Constructs a Qwen3VL processor which wraps a Qwen3VL image processor and a Qwen2 tokenizer into a single processor. [`Qwen3VLProcessor`] offers all the functionalities of [`Qwen2VLImageProcessor`] and [`Qwen2TokenizerFast`]. See the [`~Qwen3VLProcessor.__call__`] and [`~Qwen3VLProcessor.decode`] for more information. Args: image_processor ([`Qwen2VLImageProcessor`], *optional*): The image processor is a required input. tokenizer ([`Qwen2TokenizerFast`], *optional*): The tokenizer is a required input. video_processor ([`Qwen3VLVideoProcessor`], *optional*): The video processor is a required input. chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages in a chat into a tokenizable string. """ 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, chat_template=None, **kwargs): self.image_token = "<|image_pad|>" if not hasattr(tokenizer, "image_token") else tokenizer.image_token self.video_token = "<|video_pad|>" if not hasattr(tokenizer, "video_token") else tokenizer.video_token self.ts_token = "" if not hasattr(tokenizer, "ts_token") else tokenizer.ts_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.ts_token_id = ( tokenizer.ts_token_id if getattr(tokenizer, "ts_token_id", None) else tokenizer.convert_tokens_to_ids(self.ts_token) ) super().__init__(image_processor, tokenizer, video_processor, chat_template=chat_template) self.vision_start_token = ( "<|vision_start|>" if not hasattr(tokenizer, "vision_start_token") else tokenizer.vision_start_token ) self.vision_end_token = ( "<|vision_end|>" if not hasattr(tokenizer, "vision_end_token") else tokenizer.vision_end_token ) self.vision_start_token_id = ( tokenizer.vision_start_token_id if getattr(tokenizer, "vision_start_token_id", None) else tokenizer.convert_tokens_to_ids(self.vision_start_token) ) self.vision_end_token_id = ( tokenizer.vision_end_token_id if getattr(tokenizer, "vision_end_token_id", None) else tokenizer.convert_tokens_to_ids(self.vision_end_token) ) self.ts_start_token = ( "<|ts|>" if not hasattr(tokenizer, "ts_start_token") else tokenizer.ts_start_token ) self.ts_end_token = ( "<|/ts|>" if not hasattr(tokenizer, "ts_end_token") else tokenizer.ts_end_token ) self.ts_start_token_id = ( tokenizer.ts_start_token_id if getattr(tokenizer, "ts_start_token_id", None) else tokenizer.convert_tokens_to_ids(self.ts_start_token) ) self.ts_end_token_id = ( tokenizer.ts_end_token_id if getattr(tokenizer, "ts_end_token_id", None) else tokenizer.convert_tokens_to_ids(self.ts_end_token) ) def time_series_preprocessor(self,conversation): if isinstance(conversation, (list, tuple)) and ( isinstance(conversation[0], (list, tuple)) or hasattr(conversation[0], "content") ): conversations = conversation else: conversations = [conversation] batch_time_series = [] batch_time_series_metadata = [] for conversation in conversations: for message in conversation: if message['role'] != "user": continue time_series_fnames = [ content["data"] for content in message["content"] if content.get("type") == "time_series" and "data" in content ] time_series_rates = [ content.get("sampling_rate", None) for content in message["content"] if content.get("type") == "time_series" ] for path, rate in zip(time_series_fnames, time_series_rates): batch_time_series.append(path) batch_time_series_metadata.append(rate) return {"time_series_paths": batch_time_series if batch_time_series else None, "time_series_sampling_rates": batch_time_series_metadata if batch_time_series_metadata else None} def time_series_processor(self, ts_paths: list[str], sampling_rates: list[float], do_normalize=True, do_truncate=True, )-> BatchFeature: assert len(ts_paths)==len(sampling_rates), "ts_paths and sampling_rates must have the same length" ts_values=[] ts_sr=[] ts_lens=[] for idx,ts_path in enumerate(ts_paths): sr=sampling_rates[idx] ext = os.path.splitext(ts_path)[-1].lower() if ext in [".wav",'.mp3','.flac']: try: import soundfile as sf except ImportError: raise ImportError("Please install soundfile to process audio files.") ts_input, sr = sf.read(ts_path) # ts_input: np.ndarray, shape [T] or [T, C] elif ext == ".csv": pd = __import__("pandas") df = pd.read_csv(ts_path, header=None) ts_input = df.values # [T, C] elif ext == ".npy": ts_input = np.load(ts_path) # [T, C] else: raise ValueError(f"Unsupported file format: {ext}") # ts_tensor = torch.from_numpy(ts_input).float() if not isinstance(ts_input, np.ndarray): ts_input = np.array(ts_input, dtype=np.float32) if do_normalize: mean = ts_input.mean(axis=0, keepdims=True) std = ts_input.std(axis=0, keepdims=True) ts_input = (ts_input - mean) / (std + 1e-8) if do_truncate and len(ts_input)>240000: ts_input=ts_input[:240000] # truncate to 240k to avoid oom if ts_input.ndim==1: ts_input=ts_input[:, None] #[T,C] ts_len=ts_input.shape[0] if sr is None or sr == 0: # if no sr provided sr = ts_len/4 ts_values.append(ts_input) ts_sr.append(sr) ts_lens.append(ts_len) ts_lens = np.array(ts_lens) ts_sr = np.array(ts_sr) num_ts_tokens = self._get_num_ts_tokens(sampling_rates=ts_sr, ts_lens=ts_lens) return BatchFeature(data={"ts_values": ts_values, "ts_sr":ts_sr, "ts_lens":ts_lens, "num_ts_tokens":num_ts_tokens} ) def _get_num_ts_tokens(self,sampling_rates,ts_lens): strides = np.floor(160/((1+np.exp(-sampling_rates/100))**6)) patch_sizes = strides * 2 embed_lengths = (np.ceil((ts_lens - patch_sizes) / strides) + 1).astype(np.int64) num_ts_tokens=[(embed_length // 2 + 1) // 2 for embed_length in embed_lengths] return num_ts_tokens def __call__( self, images: ImageInput = None, text: Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]] = None, videos: VideoInput = None, time_series_paths: Optional[list[str]]=None, time_series_sampling_rates: Optional[list[float]]=None, **kwargs: Unpack[InternS1ProProcessorKwargs], ) -> 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 `kwrags` 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. time_series_signals (`list[np.ndarray]`, `list[torch.Tensor]`): 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`. """ output_kwargs = self._merge_kwargs( InternS1ProProcessorKwargs, tokenizer_init_kwargs=self.tokenizer.init_kwargs, **kwargs, ) if images is not None: image_inputs = self.image_processor(images=images, **output_kwargs["images_kwargs"]) image_grid_thw = image_inputs["image_grid_thw"] else: image_inputs = {} image_grid_thw = None if videos is not None: videos_inputs = self.video_processor(videos=videos, **output_kwargs["videos_kwargs"]) video_grid_thw = videos_inputs["video_grid_thw"] # If user has not requested video metadata, pop it if "return_metadata" not in kwargs: video_metadata = videos_inputs.pop("video_metadata") else: video_metadata = videos_inputs["video_metadata"] video_grid_thw = videos_inputs["video_grid_thw"] else: videos_inputs = {} video_grid_thw = None if not isinstance(text, list): text = [text] text = text.copy() # below lines change text in-place if image_grid_thw 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, "<|placeholder|>" * num_image_tokens, 1) index += 1 text[i] = text[i].replace("<|placeholder|>", self.image_token) if video_grid_thw 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]: metadata = video_metadata[index] if metadata.fps is None: logger.warning_once( "Qwen3VL requires frame timestamps to construct prompts, but the `fps` of the input video could not be inferred. " "Probably `video_metadata` was missing from inputs and you passed pre-sampled frames. " "Defaulting to `fps=24`. Please provide `video_metadata` for more accurate results." ) metadata.fps = 24 if metadata.fps is None else metadata.fps # if timestamps are not provided, calculate them curr_timestamp = self._calculate_timestamps( metadata.frames_indices, metadata.fps, self.video_processor.merge_size, ) video_placeholder = "" frame_seqlen = video_grid_thw[index][1:].prod() // merge_length for frame_idx in range(video_grid_thw[index][0]): curr_time = curr_timestamp[frame_idx] video_placeholder += f"<{curr_time:.1f} seconds>" video_placeholder += ( self.vision_start_token + "<|placeholder|>" * frame_seqlen + self.vision_end_token ) if f"{self.vision_start_token}{self.video_token}{self.vision_end_token}" in text[i]: text[i] = text[i].replace( f"{self.vision_start_token}{self.video_token}{self.vision_end_token}", video_placeholder, 1 ) else: # vllm may input video token directly text[i] = text[i].replace(self.video_token, video_placeholder, 1) index += 1 text[i] = text[i].replace("<|placeholder|>", self.video_token) time_series_inputs = {} if images is None and videos is None and time_series_paths is not None: assert time_series_sampling_rates is not None, "If time_series_signals is provided, time_series_sampling_rates must also be provided." assert len(time_series_paths) == len(time_series_sampling_rates), "The number of time series signals must match the number of sampling rates." time_series_inputs = self.time_series_processor(ts_paths=time_series_paths, sampling_rates=time_series_sampling_rates) num_ts_tokens = time_series_inputs.pop("num_ts_tokens") assert len(num_ts_tokens) == len(text), "The number of time series signals must match the number of text prompts." for i in range(len(text)): if f"{self.ts_start_token}{self.ts_token}{self.ts_end_token}" in text[i]: ts_placeholder = self.ts_start_token + self.ts_token * num_ts_tokens[i] + self.ts_end_token text[i] = text[i].replace( f"{self.ts_start_token}{self.ts_token}{self.ts_end_token}", ts_placeholder, 1 ) elif self.ts_token in text[i]: text[i] = text[i].replace(self.ts_token, self.ts_token * num_ts_tokens[i]) 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) 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,**time_series_inputs}, tensor_type=return_tensors) def _get_num_multimodal_tokens(self, image_sizes=None, video_sizes=None, **kwargs): """ Computes the number of placeholder tokens needed for multimodal inputs with the given sizes. Args: image_sizes (`list[list[int]]`, *optional*): The input sizes formatted as (height, width) per each image. video_sizes (`list[list[int]]`, *optional*): The input sizes formatted as (num_frames, height, width) per each video. Returns: `MultiModalData`: A `MultiModalData` object holding number of tokens per each of the provided input modalities, along with other useful data. """ vision_data = {} if image_sizes is not None: images_kwargs = InternS1ProProcessorKwargs._defaults.get("images_kwargs", {}) images_kwargs.update(kwargs) merge_size = images_kwargs.get("merge_size", None) or self.image_processor.merge_size num_image_patches = [ self.image_processor.get_number_of_image_patches(*image_size, images_kwargs) for image_size in image_sizes ] num_image_tokens = [(num_patches // merge_size**2) for num_patches in num_image_patches] vision_data.update({"num_image_tokens": num_image_tokens, "num_image_patches": num_image_patches}) if video_sizes is not None: videos_kwargs = InternS1ProProcessorKwargs._defaults.get("videos_kwargs", {}) videos_kwargs.update(kwargs) num_video_patches = [ self.video_processor.get_number_of_video_patches(*video_size, videos_kwargs) for video_size in video_sizes ] num_video_tokens = [(num_patches // merge_size**2) for num_patches in num_video_patches] vision_data["num_video_tokens"] = num_video_tokens return MultiModalData(**vision_data) def post_process_image_text_to_text( self, generated_outputs, skip_special_tokens=True, clean_up_tokenization_spaces=False, **kwargs ): """ Post-process the output of the model to decode the text. Args: generated_outputs (`torch.Tensor` or `np.ndarray`): The output of the model `generate` function. The output is expected to be a tensor of shape `(batch_size, sequence_length)` or `(sequence_length,)`. skip_special_tokens (`bool`, *optional*, defaults to `True`): Whether or not to remove special tokens in the output. Argument passed to the tokenizer's `batch_decode` method. clean_up_tokenization_spaces (`bool`, *optional*, defaults to `False`): Whether or not to clean up the tokenization spaces. Argument passed to the tokenizer's `batch_decode` method. **kwargs: Additional arguments to be passed to the tokenizer's `batch_decode method`. Returns: `list[str]`: The decoded text. """ return self.tokenizer.batch_decode( generated_outputs, skip_special_tokens=skip_special_tokens, clean_up_tokenization_spaces=clean_up_tokenization_spaces, **kwargs, ) def _calculate_timestamps(self, indices: Union[list[int], np.ndarray], video_fps: float, merge_size: int = 2): if not isinstance(indices, list): indices = indices.tolist() if len(indices) % merge_size != 0: indices.extend(indices[-1] for _ in range(merge_size - len(indices) % merge_size)) timestamps = [idx / video_fps for idx in indices] # @JJJYmmm frames are merged by self.merge_size, \ # so we need to average the timestamps between the first/last frame within the temporal patch timestamps = [ (timestamps[i] + timestamps[i + merge_size - 1]) / 2 for i in range(0, len(timestamps), merge_size) ] return timestamps __all__ = ["InternS1ProProcessor"]