Final Commit

Dockerfile

@@ -1,7 +1,3 @@
-# --------------------------
-# Hugging Face Space (Docker)
-# CPU-friendly default base
-# --------------------------
 FROM python:3.12-slim
 
 # System deps (add build-essential + common runtime libs for OpenCV/ONNX)

app/deps.py

@@ -40,14 +40,9 @@ def get_hf_token(
     return None
 
 def build_agent_with_token(hf_token: Optional[str]) -> SQLAgent:
-    """
-    Returns the shared SQLAgent but (temporarily) sets the token when present.
-    Avoids re-instantiating clients for every request.
-    """
-    if hf_token:
-        _sqlgen.set_token(hf_token)
-        _ansllm.set_token(hf_token)
-    else:
-        _sqlgen.set_token(settings.HF_TOKEN)
-        _ansllm.set_token(settings.HF_TOKEN)
+    tok = hf_token or settings.HF_TOKEN
+    if not tok:
+        raise RuntimeError("HF token missing: pass X-HF-Token or set HF_TOKEN")
+    _sqlgen.set_token(tok)
+    _ansllm.set_token(tok)
     return _sql_agent

app/main.py

@@ -199,4 +199,38 @@ async def query(req: QueryReq, hf_token: str | None = Depends(get_hf_token)):
             # uncertainty=0.15
         )
     except Exception as e:
-        raise HTTPException(status_code=400, detail=f"Query failed: {e}")
+        raise HTTPException(status_code=400, detail=f"Query failed: {e}")
+
+import requests
+
+@app.get("/llm/health")
+def llm_health(hf_token: str | None = Depends(get_hf_token)):
+    tok = hf_token or settings.HF_TOKEN
+    if not tok:
+        return {"status": 400, "ok": False, "body": "HF token missing"}
+
+    try:
+        r = requests.post(
+            "https://router.huggingface.co/v1/chat/completions",
+            headers={
+                "Authorization": f"Bearer {tok}",
+                "Accept": "application/json",
+                "Accept-Encoding": "identity",   # <---
+            },
+            json={
+                "model": settings.LLM_MODEL_ID,
+                "messages": [{"role":"user","content":[{"type":"text","text":"ping"}]}],
+                "max_tokens": 1,
+                "stream": False
+            },
+            timeout=20
+        )
+        ok = r.ok
+        # If OK, don't read text (keeps it light)
+        body = "ok" if ok else (r.text[:200] if r.text else str(r.content[:200]))
+        return {"status": r.status_code, "ok": ok, "body": body, "ce": r.headers.get("content-encoding")}
+    except requests.exceptions.ContentDecodingError as e:
+        # Rare gzip mismatch – report clearly
+        return {"status": 502, "ok": False, "body": f"gzip decode error: {e.__class__.__name__}"}
+    except Exception as e:
+        return {"status": 500, "ok": False, "body": f"{type(e).__name__}: {e}"}

app/tools/llm_answer.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 from typing import Optional, Dict, Any, List
 import requests, json
 
-HF_CHAT_URL = "https://router.huggingface.co/featherless-ai/v1/chat/completions"
+HF_CHAT_URL = "https://router.huggingface.co/v1/chat/completions"
 
 SYSTEM_PROMPT = """You are a BI copilot.
 - NEVER invent numbers; only summarize from provided table rows.
@@ -44,7 +44,10 @@ class AnswerLLM:
             "temperature": self.temperature,
             "max_tokens": self.max_tokens,
         }
-        headers = {"Authorization": f"Bearer {self.token}"}
+        headers = {"Authorization": f"Bearer {self.token}",
+                    "Accept": "application/json",
+                    "Accept-Encoding": "identity",
+        }
         r = requests.post(HF_CHAT_URL, headers=headers, json=payload, timeout=self.timeout)
         r.raise_for_status()
         return r.json()["choices"][0]["message"]["content"]

app/tools/llm_sqlgen.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 from typing import Optional, Dict, Any
 import requests, json
 
-HF_CHAT_URL = "https://router.huggingface.co/featherless-ai/v1/chat/completions"
+HF_CHAT_URL = "https://router.huggingface.co/v1/chat/completions"
 
 SCHEMA_SPEC = """
 Tables and columns (SQLite):
@@ -24,6 +24,10 @@ Rules:
 - For monthly rollups use strftime('%Y-%m', day).
 - Join to dim_product when you need category/name/price.
 - For per-employee metrics use fact_sales_detail (employee_id may be NULL for Online).
+- Always generate the SQL Queries in English 
+for example. 
+    "q": रमेश का टोटल जेनरेटेड रेवेन्यू बताओ 
+    "sql": SELECT SUM(d.revenue) AS total_revenue FROM fact_sales_detail d JOIN dim_employee e ON e.emp_id = d.employee_id WHERE e.name LIKE 'Ramesh %'
 """
 
 FEW_SHOTS = [
@@ -93,7 +97,10 @@ class SQLGenTool:
             "temperature": self.temperature,
             "max_tokens": self.max_tokens,
         }
-        headers = {"Authorization": f"Bearer {self.token}"}
+        headers = {"Authorization": f"Bearer {self.token}",
+                    "Accept": "application/json",
+                    "Accept-Encoding": "identity"
+                }
         r = requests.post(HF_CHAT_URL, headers=headers, json=payload, timeout=self.timeout)
         r.raise_for_status()
         content = r.json()["choices"][0]["message"]["content"].strip()

requirements.txt

@@ -5,7 +5,6 @@ pydantic-settings
 numpy==1.26.4
 faiss-cpu
 insightface==0.7.3
-# onnxruntime
 onnxruntime==1.17.3
 opencv-python==4.10.0.84
 python-multipart