Update app.py

app.py

@@ -5,7 +5,6 @@ import time
 import subprocess
 import tempfile
 from pathlib import Path
-from huggingface_hub import InferenceClient
 
 # ── INSTALL ───────────────────────────────────────────────────────────────────
 for pkg in ["pypdf", "edge-tts", "pydub", "pillow", "pytesseract", "openai-whisper", "ffmpeg-python"]:
@@ -19,27 +18,61 @@ from pydub import AudioSegment
 from PIL import Image
 import pytesseract
 
-# ── MODEL — free on HF Pro, no credit cost ───────────────────────────────────
-# Primary: Mistral 7B — free, fast, no credit drain
-# Fallback: Zephyr 7B — also free
-PRIMARY_MODEL  = "mistralai/Mistral-7B-Instruct-v0.3"
-FALLBACK_MODEL = "HuggingFaceH4/zephyr-7b-beta"
+# ── MODEL — FREE HF INFERENCE API ────────────────────────────────────────────
+# Bypasses router.huggingface.co (PAID) entirely.
+# Uses api-inference.huggingface.co — completely free, same HF_TOKEN, no credits.
+import requests as _req
+
+_FREE_API = "https://api-inference.huggingface.co/models/{model}"
+_MODELS = [
+    "HuggingFaceH4/zephyr-7b-beta",
+    "mistralai/Mistral-7B-Instruct-v0.3",
+    "microsoft/Phi-3-mini-4k-instruct",
+]
 
 def call_llama(prompt: str, max_tokens: int = 2000) -> str:
     token = os.environ.get("HF_TOKEN", "")
-    for model in [PRIMARY_MODEL, FALLBACK_MODEL]:
+    headers = {"Authorization": f"Bearer {token}"} if token else {}
+
+    for model in _MODELS:
         try:
-            client = InferenceClient(model=model, token=token if token else None)
-            response = client.chat_completion(
-                messages=[{"role": "user", "content": prompt}],
-                max_tokens=max_tokens,
-                temperature=0.7,
-            )
-            return response.choices[0].message.content
-        except Exception as e:
-            last_err = str(e)
+            # Format prompt correctly for each model family
+            if "zephyr" in model:
+                fmt = f"<|system|>\nYou are a precise legal document analyst.</s>\n<|user|>\n{prompt}</s>\n<|assistant|>\n"
+            elif "Phi-3" in model:
+                fmt = f"<|user|>\n{prompt}<|end|>\n<|assistant|>\n"
+            else:
+                fmt = f"[INST] {prompt} [/INST]"
+
+            def _post(f):
+                return _req.post(
+                    _FREE_API.format(model=model), headers=headers,
+                    json={"inputs": f,
+                          "parameters": {"max_new_tokens": max_tokens,
+                                         "temperature": 0.7,
+                                         "return_full_text": False,
+                                         "do_sample": True}},
+                    timeout=120
+                )
+
+            resp = _post(fmt)
+
+            # Model loading — wait and retry once
+            if resp.status_code == 503:
+                import time; time.sleep(20)
+                resp = _post(fmt)
+
+            if resp.status_code == 200:
+                result = resp.json()
+                if isinstance(result, list) and result:
+                    text = result[0].get("generated_text", "").strip()
+                    if text:
+                        return text
+
+        except Exception:
             continue
-    return f"Could not reach AI — check your HF_TOKEN is set in Space secrets. ({last_err})"
+
+    return "Could not reach AI. Make sure HF_TOKEN is set in your Space secrets."
 
 
 # ── FILE EXTRACTION ───────────────────────────────────────────────────────────