File size: 3,256 Bytes
34a3ef8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from typing import List, Dict, Any
import uvicorn
from model_loader import get_local_llm_instance

app = FastAPI(title="Stateless Agent Pipeline")

# Enable global cross-origin resource sharing for frontend html access
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Load model engine universally on runtime startup
try:
    llm_instance = get_local_llm_instance()
except Exception as init_err:
    print(f"[CRITICAL ERROR] Failed to load local weights: {init_err}")
    llm_instance = None

# Validation structure for parsing the data packets cleanly
class ChatPayload(BaseModel):
    user_id: str
    user_message: str
    current_chat_history: List[Dict[str, Any]] = []
    user_files: Dict[str, Any] = {}

@app.get("/")
def read_root():
    return {"status": "online", "engine": "Llama.cpp local cluster running flawlessly"}

@app.post("/chat")
async def chat_endpoint(payload: ChatPayload):
    global llm_instance
    if llm_instance is None:
        raise HTTPException(status_code=500, detail="Local LLM instance cluster is offline.")
    
    try:
        user_query = payload.user_message
        
        # Build strict system directives for clean output responses
        system_instruction = (
            "<|im_start|>system\n"
            "You are a helpful, extremely fast AI assistant. "
            "Respond cleanly, accurately and directly to the prompt. "
            "Keep formatting minimal.<|im_end|>\n"
        )
        
        # Format chat history context string if it exists
        history_context = ""
        for turn in payload.current_chat_history[-4:]: # Keep only the last 4 exchanges to preserve fast RAM context
            role = "user" if turn.get("role") == "user" else "assistant"
            content = turn.get("content", "")
            history_context += f"<|im_start|>{role}\n{content}<|im_end|>\n"
            
        # Compile complete operational template string
        final_prompt = f"{system_instruction}{history_context}<|im_start|>user\n{user_query}<|im_end|>\n<|im_start|>assistant\n"
        
        # Run synchronous inference across CPU matrix
        output = llm_instance(
            final_prompt,
            max_tokens=512,       # Generation constraint for faster response times
            stop=["<|im_end|>", "<|im_start|>", "user:", "assistant:"],
            echo=False
        )
        
        generated_text = output["choices"][0]["text"].strip()
        
        # Re-construct updated structural array history block
        updated_history = payload.current_chat_history + [
            {"role": "user", "content": user_query},
            {"role": "assistant", "content": generated_text}
        ]
        
        return {
            "updated_chat_history": updated_history,
            "updated_files": payload.user_files
        }
        
    except Exception as exec_error:
        raise HTTPException(status_code=500, detail=f"Inference Engine Error: {str(exec_error)}")

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)