-
-
Notifications
You must be signed in to change notification settings - Fork 16
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
Showing
5 changed files
with
234 additions
and
10 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
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
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 +1,181 @@ | ||
| try: | ||
| from deepseek_vl.models import VLChatProcessor | ||
| except: | ||
| VLChatProcessor = None | ||
| from transformers import AutoModelForCausalLM | ||
| from datetime import datetime | ||
| import requests | ||
| import torch | ||
| import PIL.Image | ||
| import uuid | ||
| import os | ||
| import base64 | ||
| from ezlocalai.Helpers import get_tokens | ||
|
|
||
|
|
||
| class VLM: | ||
| def __init__(self, model="deepseek-ai/deepseek-vl-1.3b-chat"): | ||
| self.model = model.split("/")[-1] | ||
| self.params = {} | ||
| os.makedirs(os.path.join(os.getcwd(), "outputs"), exist_ok=True) | ||
| try: | ||
| self.vl_chat_processor = VLChatProcessor.from_pretrained(model) | ||
| self.tokenizer = self.vl_chat_processor.tokenizer | ||
| self.vl_gpt = AutoModelForCausalLM.from_pretrained( | ||
| model, | ||
| trust_remote_code=True, | ||
| cache_dir=os.path.join(os.getcwd(), "models"), | ||
| ) | ||
| if torch.cuda.is_available(): | ||
| self.vl_gpt = self.vl_gpt.to(torch.bfloat16).cuda().eval() | ||
| else: | ||
| self.vl_gpt = self.vl_gpt.to(torch.bfloat16).eval() | ||
| except Exception as e: | ||
| print(f"[VLM] Error: {e}") | ||
| self.vl_chat_processor = None | ||
| self.tokenizer = None | ||
| self.vl_gpt = None | ||
|
|
||
| def chat(self, messages, **kwargs): | ||
| pil_images = [] | ||
| images = [] | ||
| prompt = "" | ||
| for message in messages: | ||
| if isinstance(message["content"], str): | ||
| role = message["role"] if "role" in message else "User" | ||
| if role.lower() == "user": | ||
| prompt += f"{message['content']}\n\n" | ||
| if role.lower() == "system": | ||
| prompt = f"System: {message['content']}\n\nUser: {prompt}" | ||
| if isinstance(message["content"], list): | ||
| for msg in message["content"]: | ||
| if "text" in msg: | ||
| role = message["role"] if "role" in message else "User" | ||
| if role.lower() == "user": | ||
| prompt += f"{msg['text']}\n\n" | ||
| if "image_url" in msg: | ||
| url = str( | ||
| msg["image_url"]["url"] | ||
| if "url" in msg["image_url"] | ||
| else msg["image_url"] | ||
| ) | ||
| image_path = f"./outputs/{uuid.uuid4().hex}.jpg" | ||
| if url.startswith("http"): | ||
| image = requests.get(url).content | ||
| else: | ||
| file_type = url.split(",")[0].split("/")[1].split(";")[0] | ||
| if file_type == "jpeg": | ||
| file_type = "jpg" | ||
| image_path = f"./outputs/{uuid.uuid4().hex}.{file_type}" | ||
| if "," in url: | ||
| image = base64.b64decode(url.split(",")[1]) | ||
| else: | ||
| image = base64.b64decode(url) | ||
| with open(image_path, "wb") as f: | ||
| f.write(image) | ||
| images.append(image_path) | ||
| pil_img = PIL.Image.open(image_path) | ||
| pil_img = pil_img.convert("RGB") | ||
| pil_images.append(pil_img) | ||
| if len(images) > 0: | ||
| for image in images: | ||
| prompt = f"<image_placeholder> {prompt}" | ||
| conversation = [ | ||
| {"role": "User", "content": prompt, "images": images}, | ||
| {"role": "Assistant", "content": ""}, | ||
| ] | ||
| prepare_inputs = self.vl_chat_processor( | ||
| conversations=conversation, images=pil_images, force_batchify=True | ||
| ).to(self.vl_gpt.device) | ||
| inputs_embeds = self.vl_gpt.prepare_inputs_embeds(**prepare_inputs) | ||
| outputs = self.vl_gpt.language_model.generate( | ||
| inputs_embeds=inputs_embeds, | ||
| attention_mask=prepare_inputs.attention_mask, | ||
| pad_token_id=self.tokenizer.eos_token_id, | ||
| bos_token_id=self.tokenizer.bos_token_id, | ||
| eos_token_id=self.tokenizer.eos_token_id, | ||
| max_new_tokens=1024, | ||
| do_sample=False, | ||
| use_cache=True, | ||
| ) | ||
| answer = self.tokenizer.decode( | ||
| outputs[0].cpu().tolist(), skip_special_tokens=True | ||
| ) | ||
| completion_tokens = get_tokens(answer) | ||
| prompt_tokens = get_tokens( | ||
| " ".join([message["content"] for message in conversation]) | ||
| ) | ||
| total_tokens = completion_tokens + prompt_tokens | ||
| data = { | ||
| "choices": [ | ||
| { | ||
| "finish_reason": "stop", | ||
| "index": 0, | ||
| "message": {"content": answer, "role": "assistant"}, | ||
| "logprobs": None, | ||
| } | ||
| ], | ||
| "created": datetime.now().isoformat(), | ||
| "id": f"chatcmpl-{uuid.uuid4().hex}", | ||
| "model": self.model, | ||
| "object": "chat.completion", | ||
| "usage": { | ||
| "completion_tokens": completion_tokens, | ||
| "prompt_tokens": prompt_tokens, | ||
| "total_tokens": total_tokens, | ||
| }, | ||
| } | ||
| return data | ||
|
|
||
| def completion(self, prompt, **kwargs): | ||
| messages = [ | ||
| {"role": "User", "content": prompt}, | ||
| ] | ||
| completion = self.chat( | ||
| messages=messages, | ||
| max_tokens=kwargs["max_tokens"] if "max_tokens" in kwargs else 1024, | ||
| ) | ||
| data = { | ||
| "choices": [ | ||
| { | ||
| "finish_reason": "length", | ||
| "index": 0, | ||
| "logprobs": None, | ||
| "text": completion["choices"][0]["message"]["content"], | ||
| } | ||
| ], | ||
| "created": datetime.now().isoformat(), | ||
| "id": f"cmpl-{uuid.uuid4().hex}", | ||
| "model": self.model, | ||
| "object": "text_completion", | ||
| "usage": { | ||
| "completion_tokens": completion["usage"]["completion_tokens"], | ||
| "prompt_tokens": completion["usage"]["prompt_tokens"], | ||
| "total_tokens": completion["usage"]["total_tokens"], | ||
| }, | ||
| } | ||
| return data | ||
|
|
||
| def describe_image(self, image_url): | ||
| messages = [ | ||
| { | ||
| "role": "User", | ||
| "content": [ | ||
| {"type": "image_url", "image_url": {"url": image_url}}, | ||
| { | ||
| "type": "text", | ||
| "text": "Describe each stage of this image.", | ||
| }, | ||
| ], | ||
| }, | ||
| ] | ||
| response = self.chat( | ||
| messages=messages, | ||
| ) | ||
| return response["choices"][0]["message"]["content"] | ||
|
|
||
| def models(self): | ||
| return [ | ||
| "deepseek-ai/deepseek-vl-1.3b-chat", | ||
| "deepseek-ai/deepseek-vl-7b-chat", | ||
| ] |