Spaces:
Running
on
Zero
Running
on
Zero
File size: 3,801 Bytes
f04732f a3db70a f04732f 08bc998 a3db70a 4be8019 a3db70a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 |
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread
import re
import time
from PIL import Image
import torch
import spaces
import subprocess
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
torch.set_default_device('cuda')
tokenizer = AutoTokenizer.from_pretrained(
'qnguyen3/nanoLLaVA',
trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
'qnguyen3/nanoLLaVA',
torch_dtype=torch.float16,
device_map='auto',
trust_remote_code=True)
@spaces.GPU
def bot_streaming(message, history):
messages = []
if message["files"]:
image = message["files"][-1]["path"]
else:
for i, hist in enumerate(history):
if type(hist[0])==tuple:
image = hist[0][0]
image_turn = i
if len(history) > 0 and image is not None:
messages.append({"role": "user", "content": f'<image>\n{history[1][0]}'})
messages.append({"role": "assistant", "content": history[1][1] })
for human, assistant in history[2:]:
messages.append({"role": "user", "content": human })
messages.append({"role": "assistant", "content": assistant })
messages.append({"role": "user", "content": message['text']})
elif len(history) > 0 and image is None:
for human, assistant in history:
messages.append({"role": "user", "content": human })
messages.append({"role": "assistant", "content": assistant })
messages.append({"role": "user", "content": message['text']})
elif len(history) == 0 and image is not None:
messages.append({"role": "user", "content": f"<image>\n{message['text']}"})
elif len(history) == 0 and image is None:
messages.append({"role": "user", "content": message['text'] })
# if image is None:
# gr.Error("You need to upload an image for LLaVA to work.")
image = Image.open(image).convert("RGB")
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True)
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0)
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens = True)
image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)
generation_kwargs = dict(input_ids=input_ids, images=image_tensor, streamer=streamer, max_new_tokens=100)
generated_text = ""
thread = Thread(target=model.generate, kwargs=generation_kwargs)
thread.start()
text_prompt =f"<|im_start|>user\n{message['text']}<|im_end|>"
buffer = ""
for new_text in streamer:
buffer += new_text
generated_text_without_prompt = buffer[len(text_prompt):]
time.sleep(0.04)
yield generated_text_without_prompt
demo = gr.ChatInterface(fn=bot_streaming, title="LLaVA NeXT", examples=[{"text": "What is on the flower?", "files":["./bee.jpg"]},
{"text": "How to make this pastry?", "files":["./baklava.png"]}],
description="Try [LLaVA NeXT](https://huggingface.co/docs/transformers/main/en/model_doc/llava_next) in this demo (more specifically, the [Mistral-7B variant](https://huggingface.co/llava-hf/llava-v1.6-mistral-7b-hf)). Upload an image and start chatting about it, or simply try one of the examples below. If you don't upload an image, you will receive an error.",
stop_btn="Stop Generation", multimodal=True)
demo.launch(debug=True) |