-
Notifications
You must be signed in to change notification settings - Fork 18
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
matatonic
committed
Sep 13, 2024
1 parent
c615dcc
commit e438d38
Showing
8 changed files
with
211 additions
and
126 deletions.
There are no files selected for viewing
This file contains 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 |
|---|---|---|
|
|
@@ -5,6 +5,7 @@ on: | |
| push: | ||
| branches: | ||
| - 'main' | ||
| - 'dev' | ||
| release: | ||
| types: [published] | ||
|
|
||
|
|
||
This file contains 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 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 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 +1,83 @@ | ||
| from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor | ||
| from qwen_vl_utils import process_vision_info | ||
|
|
||
| import os | ||
| from vision_qna import * | ||
|
|
||
| # Qwen/Qwen2-VL-2B-Instruct-AWQ | ||
| # Qwen/Qwen2-VL-2B-Instruct | ||
| # Qwen/Qwen2-VL-7B-Instruct-AWQ | ||
| # Qwen/Qwen2-VL-7B-Instruct | ||
| # X Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int4 | ||
| # X Qwen/Qwen2-VL-2B-Instruct-GPTQ-Int8 | ||
| # X Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int4 | ||
| # X Qwen/Qwen2-VL-7B-Instruct-GPTQ-Int8 | ||
|
|
||
| class VisionQnA(VisionQnABase): | ||
| model_name: str = "qwen2-vl" | ||
| format: 'chatml' | ||
| vision_layers: List[str] = ['visual'] | ||
|
|
||
| def __init__(self, model_id: str, device: str, device_map: str = 'auto', extra_params = {}, format = None): | ||
| super().__init__(model_id, device, device_map, extra_params, format) | ||
|
|
||
| self.processor = AutoProcessor.from_pretrained(model_id) | ||
|
|
||
| del self.params['trust_remote_code'] | ||
| self.model = Qwen2VLForConditionalGeneration.from_pretrained(**self.params).eval() | ||
|
|
||
| self.loaded_banner() | ||
|
|
||
| async def stream_chat_with_images(self, request: ImageChatRequest) -> AsyncGenerator[str, None]: | ||
| # image_tag = '<|vision_start|><|image_pad|><|vision_end|>' | ||
|
|
||
| messages = [] | ||
|
|
||
| for m in request.messages: | ||
| if m.role == 'user': | ||
| msg = { 'role': m.role, 'content': [] } | ||
| for c in m.content: | ||
| if c.type == 'image_url': | ||
| # hack around https://github.com/QwenLM/Qwen2-VL/issues/202' | ||
| if c.image_url.url.startswith('data:image'): | ||
| parts = c.image_url.url.split(';') | ||
| if parts[1].startswith('charset='): | ||
| c.image_url.url = parts[0] + ';' + parts[2] | ||
|
|
||
| msg['content'].extend([{'type': c.type, 'image': c.image_url.url}]) | ||
| elif c.type == 'text': | ||
| msg['content'].extend([{'type': c.type, 'text': c.text}]) | ||
| elif c.type == 'video': # not likely to work. | ||
| msg['content'].extend([{'type': c.type, 'video': c.image_url.url}]) | ||
| else: | ||
| #msg = { 'role': m.role, 'content': [{ 'type': 'text', 'text': c.text }] } | ||
| msg = { 'role': m.role, 'content': c.text } | ||
|
|
||
| messages.extend([msg]) | ||
|
|
||
| text = self.processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) | ||
|
|
||
| image_inputs, video_inputs = process_vision_info(messages) | ||
| inputs = self.processor( | ||
| text=[text], | ||
| images=image_inputs, | ||
| videos=video_inputs, | ||
| padding=True, | ||
| return_tensors="pt", | ||
| ) | ||
| inputs = inputs.to(self.device) | ||
|
|
||
| params = self.get_generation_params(request, default_params={}) | ||
|
|
||
| generation_kwargs = dict( | ||
| **inputs, | ||
| **params, | ||
| ) | ||
|
|
||
| for new_text in threaded_streaming_generator(generate=self.model.generate, tokenizer=self.processor.tokenizer, generation_kwargs=generation_kwargs): | ||
| end = new_text.find(self.processor.tokenizer.eos_token) | ||
| if end == -1: | ||
| yield new_text | ||
| else: | ||
| yield new_text[:end] | ||
| break |
This file contains 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 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 |
|---|---|---|
|
|
@@ -41,6 +41,7 @@ matplotlib | |
| optimum | ||
| tiktoken | ||
| transformers_stream_generator | ||
| qwen-vl-utils | ||
|
|
||
| # video | ||
| decord | ||
|
|
||
Oops, something went wrong.