File size: 2,649 Bytes
0218c20
03991d8
 
 
0218c20
 
29331bd
6a547e4
29331bd
2ba12d8
6f6ae2a
 
 
 
 
2ba12d8
6f6ae2a
2ba12d8
29331bd
2ba12d8
0218c20
2ba12d8
 
 
 
 
 
 
 
 
 
 
0218c20
2ba12d8
6a547e4
0218c20
03991d8
2ba12d8
03991d8
 
2ba12d8
03991d8
 
0218c20
03991d8
 
 
 
 
2ba12d8
03991d8
 
 
2ba12d8
0218c20
 
03991d8
2ba12d8
 
 
 
 
 
 
 
 
 
 
0218c20
2ba12d8
 
0218c20
 
 
2ba12d8
0218c20
 
 
 
 
2ba12d8
 
 
 
 
 
 
 
 
6a547e4
2ba12d8
03991d8
 
 
2ba12d8
 
 
 
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
87
88
89
90
91
92
93
94
95
96
97
98
from fastapi import FastAPI
from pydantic import BaseModel
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import os
import asyncio

# Set cache directories
cache_dir = "/tmp/hf_home"
os.environ["HF_HOME"] = cache_dir
os.environ["TRANSFORMERS_CACHE"] = cache_dir
os.environ["HUGGINGFACE_HUB_CACHE"] = cache_dir

# Create cache directory with proper permissions
os.makedirs(cache_dir, exist_ok=True)
os.chmod(cache_dir, 0o777)

# Load model and tokenizer
model_name = "Qwen/Qwen2.5-0.5B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(
    model_name,
    trust_remote_code=True,
    cache_dir=cache_dir
)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    trust_remote_code=True,
    cache_dir=cache_dir,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
)

# Set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Initialize FastAPI
app = FastAPI()

# Enable CORS
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Input model
class Question(BaseModel):
    question: str

# System prompt
SYSTEM_PROMPT = "You are Orion, an intelligent AI assistant created by Abdullah Ali, a 13-year-old from Lahore. Respond kindly and wisely."

async def generate_response_chunks(prompt: str):
    # Create the chat template
    messages = [
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": prompt}
    ]
    
    # Apply chat template
    qwen_prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    
    # Tokenize and generate
    inputs = tokenizer(qwen_prompt, return_tensors="pt").to(device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=512,
        do_sample=True,
        temperature=0.7,
        top_p=0.9,
        pad_token_id=tokenizer.eos_token_id
    )
    
    # Decode and clean the output
    full_output = tokenizer.decode(outputs[0], skip_special_tokens=False)
    
    # Extract only the assistant's response
    response = full_output[len(qwen_prompt):].split(tokenizer.eos_token)[0].strip()
    
    # Stream the response
    for word in response.split():
        yield word + " "
        await asyncio.sleep(0.05)

@app.post("/ask")
async def ask(question: Question):
    return StreamingResponse(
        generate_response_chunks(question.question),
        media_type="text/plain"
    )