-
Notifications
You must be signed in to change notification settings - Fork 205
[MODEL] support deepseek_vl
#2942
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
Merged
Changes from 1 commit
Commits
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,144 @@ | ||
| # SPDX-FileCopyrightText: 2026 ModelCloud.ai | ||
| # SPDX-FileCopyrightText: 2026 qubitium@modelcloud.ai | ||
| # SPDX-License-Identifier: Apache-2.0 | ||
| # Contact: qubitium@modelcloud.ai, x.com/qubitium | ||
|
|
||
| from types import SimpleNamespace | ||
| from typing import Any, Dict | ||
|
|
||
| import torch | ||
| from transformers import AutoModelForImageTextToText, AutoProcessor, ProcessorMixin | ||
|
|
||
| from ...utils.calibration import batched | ||
| from ...utils.model import MODALITY, move_to | ||
| from ...utils.offload import offload_to_disk | ||
| from .._const import CPU | ||
| from ..base import BaseQModel | ||
|
|
||
|
|
||
| class DeepSeekVLQModel(BaseQModel): | ||
| loader = AutoModelForImageTextToText | ||
| require_load_processor = True | ||
| support_batch_quantize = False | ||
|
|
||
| modality = [MODALITY.TEXT, MODALITY.IMAGE_TO_TEXT] | ||
|
|
||
| pre_lm_head_norm_module = "model.language_model.norm" | ||
| rotary_embedding = "model.language_model.rotary_emb" | ||
| # layer_modules_strict = False | ||
|
|
||
| # HF_CONVERSION_MAP_REVERSED = ( | ||
| # # DeepSeek-VL mounts `SiglipVisionModel` at `model.vision_model`. | ||
| # # Runtime shell paths expose `model.vision_model.*`, while the original | ||
| # # checkpoint keeps SigLIP's base prefix under `model.vision_model.vision_model.*`. | ||
| # SimpleNamespace( | ||
| # source_patterns=[r"^model\.vision_model\.(?!vision_model\.)(.+)$"], | ||
| # target_patterns=[r"^model.vision_model.vision_model.\1"], | ||
| # operations=[], | ||
| # ), | ||
| # ) | ||
|
|
||
| module_tree = [ | ||
| "model", | ||
| "language_model", | ||
| "layers", | ||
| "#", | ||
| { | ||
| "input_layernorm": ("input_layernorm:!",), | ||
| "self_attn": ("q_proj:0", "k_proj:0", "v_proj:0", "o_proj:1"), | ||
| "post_attention_layernorm": ("post_attention_layernorm:!",), | ||
| "mlp": ("gate_proj:0", "up_proj:0", "down_proj:1"), | ||
| }, | ||
| ] | ||
|
|
||
| def pre_quantize_generate_hook_start(self): | ||
| core_model = self.model.model | ||
| language_model = core_model.language_model | ||
| self.shell_module_materialize(language_model.embed_tokens, self.quantize_config.device) | ||
| self.shell_module_materialize(language_model.norm, self.quantize_config.device) | ||
| self.shell_module_materialize(language_model.rotary_emb, self.quantize_config.device) | ||
| self.shell_module_materialize(core_model.vision_model, self.quantize_config.device) | ||
| self.shell_module_materialize(core_model.aligner, self.quantize_config.device) | ||
|
|
||
| def pre_quantize_generate_hook_end(self): | ||
| core_model = self.model.model | ||
| language_model = core_model.language_model | ||
| if self.quantize_config.offload_to_disk: | ||
| offload_to_disk( | ||
| model=language_model, | ||
| module=language_model.embed_tokens, | ||
| disk_path=self.quantize_config.offload_to_disk_path, | ||
| ) | ||
| offload_to_disk( | ||
| model=language_model, | ||
| module=language_model.norm, | ||
| disk_path=self.quantize_config.offload_to_disk_path, | ||
| ) | ||
| offload_to_disk( | ||
| model=language_model, | ||
| module=language_model.rotary_emb, | ||
| disk_path=self.quantize_config.offload_to_disk_path, | ||
| ) | ||
| offload_to_disk( | ||
| model=core_model, | ||
| module=core_model.vision_model, | ||
| disk_path=self.quantize_config.offload_to_disk_path, | ||
| ) | ||
| offload_to_disk( | ||
| model=core_model, | ||
| module=core_model.aligner, | ||
| disk_path=self.quantize_config.offload_to_disk_path, | ||
| ) | ||
| return | ||
|
|
||
| language_model.embed_tokens = move_to(language_model.embed_tokens, device=CPU) | ||
| language_model.norm = move_to(language_model.norm, device=CPU) | ||
| language_model.rotary_emb = move_to(language_model.rotary_emb, device=CPU) | ||
| core_model.vision_model = move_to(core_model.vision_model, device=CPU) | ||
| core_model.aligner = move_to(core_model.aligner, device=CPU) | ||
|
|
||
| def preprocess_dataset(self, sample: Dict) -> Dict: | ||
| return sample | ||
|
|
||
| def load_processor(self) -> ProcessorMixin: | ||
| return AutoProcessor.from_pretrained(self.model_local_path, trust_remote_code=False) | ||
|
|
||
| @classmethod | ||
| def prepare_inputs_for_conversations( | ||
| cls, | ||
| processor: ProcessorMixin, | ||
| conversations: list[dict] | list[list[dict]], | ||
| ): | ||
| return processor.apply_chat_template( | ||
| conversations, | ||
| add_generation_prompt=True, | ||
| tokenize=True, | ||
| return_dict=True, | ||
| return_tensors="pt", | ||
| ) | ||
|
|
||
| def prepare_dataset(self, calibration_dataset, batch_size: int = 1, **kwargs): | ||
| del kwargs | ||
| processor = self.load_processor() | ||
| calib_data = [] | ||
| for batch in batched(calibration_dataset, batch_size, process_func=self.preprocess_dataset): | ||
| calib_data.append(self.prepare_inputs_for_conversations(processor, batch)) | ||
| del processor | ||
| return calib_data | ||
|
|
||
| def move_input_capture_example(self, example: Dict[str, Any], data_device: torch.device) -> Dict[str, Any]: | ||
| example = super().move_input_capture_example(example, data_device) | ||
| pixel_values = example.get("pixel_values") | ||
| if torch.is_tensor(pixel_values): | ||
| vision_model = self.model.model.vision_model | ||
| first_parameter = next(vision_model.parameters(), None) | ||
| vision_device = getattr(first_parameter, "device", pixel_values.device) | ||
| vision_dtype = getattr(vision_model, "dtype", None) | ||
| if not isinstance(vision_dtype, torch.dtype): | ||
| vision_dtype = getattr(first_parameter, "dtype", None) | ||
| if isinstance(vision_dtype, torch.dtype): | ||
| example["pixel_values"] = pixel_values.to(device=vision_device, dtype=vision_dtype) | ||
| return example | ||
|
|
||
|
|
||
| __all__ = ["DeepSeekVLQModel"] | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.