Update app.py

app.py

@@ -1,32 +1,42 @@
-from fastapi import FastAPI, HTTPException
+ from fastapi import FastAPI, HTTPException
 from pydantic import BaseModel
 from fastapi.middleware.cors import CORSMiddleware
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
-import multiprocessing
 
 # =========================================================
-# FAST MODEL (SUB-SECOND RESPONSE)
+# MODEL (ULTRA FAST FREE-TIER)
 # =========================================================
 
-REPO_ID = "bartowski/Qwen2.5-1.5B-Instruct-GGUF"
-FILENAME = "Qwen2.5-1.5B-Instruct-Q4_K_M.gguf"
+REPO_ID = "TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF"
+FILENAME = "tinyllama-1.1b-chat.Q4_K_M.gguf"
 
-MODEL_PATH = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)
+print("[SYSTEM] Downloading TinyLlama...")
+
+MODEL_PATH = hf_hub_download(
+    repo_id=REPO_ID,
+    filename=FILENAME
+)
+
+print("[SYSTEM] Initializing model...")
 
 llm = Llama(
     model_path=MODEL_PATH,
-    n_ctx=1536,
-    n_batch=512,
-    n_threads=multiprocessing.cpu_count(),
+    n_ctx=512,           # КРИТИЧНО для скорости
+    n_batch=1024,
+    n_threads=2,         # РОВНО под HF CPU Basic
+    use_mmap=True,
+    use_mlock=False,
     verbose=False
 )
 
+print("[SYSTEM] TinyLlama READY")
+
 # =========================================================
 # FASTAPI
 # =========================================================
 
-app = FastAPI(title="Apex Fast Engine")
+app = FastAPI(title="Apex Free Engine")
 
 app.add_middleware(
     CORSMiddleware,
@@ -35,45 +45,66 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
+# =========================================================
+# REQUEST MODEL
+# =========================================================
+
 class AnalysisRequest(BaseModel):
     context: str
     query: str
 
+# =========================================================
+# ROUTES
+# =========================================================
+
 @app.get("/")
 def health():
-    return {"status": "online", "mode": "ultra-fast"}
+    return {
+        "status": "online",
+        "engine": "Apex",
+        "model": "TinyLlama-1.1B",
+        "tier": "HF Free"
+    }
 
 @app.post("/analyze")
 def analyze(req: AnalysisRequest):
     try:
-        prompt = f"""<|im_start|>system
+        # СУПЕР КОРОТКИЙ, НО УМНЫЙ PROMPT
+        prompt = f"""<|system|>
 Ты — Apex.
-Отвечай КОРОТКО, ТОЧНО, БЕЗ ВОДЫ.
-Если данных нет — скажи "Недостаточно данных".
+Отвечай кратко, логично и по делу.
+Используй только данный контекст.
+Если данных недостаточно — скажи об этом.
 Язык: русский.
-<|im_end|>
-<|im_start|>user
+</s>
+<|user|>
 Контекст:
 {req.context}
 
 Вопрос:
 {req.query}
-<|im_end|>
-<|im_start|>assistant
+</s>
+<|assistant|>
 """
 
         output = llm(
             prompt,
-            max_tokens=180,
-            temperature=0.1,
-            top_p=0.9,
-            stop=["<|im_end|>"],
+            max_tokens=60,        # БОЛЬШЕ НЕЛЬЗЯ НА FREE
+            temperature=0.1,     # Минимум фантазии
+            top_p=0.8,
+            stop=["</s>"],
             echo=False
         )
 
+        answer = output["choices"][0]["text"].strip()
+
         return {
-            "result": output["choices"][0]["text"].strip()
+            "result": answer,
+            "model": "TinyLlama-1.1B"
         }
 
     except Exception as e:
-        raise HTTPException(status_code=500, detail=str(e))
+        raise HTTPException(
+            status_code=500,
+            detail=str(e)
+        )