Update app.py

app.py

@@ -3,12 +3,6 @@
 #   PROVIDER=hf_model (default) -> calls HF Inference API for K2 (recommended for demo)
 #   PROVIDER=local              -> loads model with transformers (requires GPU Space)
 #   PROVIDER=stub               -> offline canned answers
-#
-# Space Secrets / Variables to set:
-#   HF_TOKEN        -> your Hugging Face token (Read + Inference permissions)
-#   MODEL_ID        -> default: MBZUAI-IFM/K2-Think-SFT (fallback: LLM360/K2-Think)
-#   PROVIDER        -> "hf_model" | "local" | "stub"
-#   HF_HUB_DISABLE_TELEMETRY=1 (optional)
 
 import os, time, json, random
 import requests
@@ -19,7 +13,7 @@ PROVIDER = os.getenv("PROVIDER", "hf_model").strip()
 MODEL_ID = os.getenv("MODEL_ID", "MBZUAI-IFM/K2-Think-SFT").strip()
 HF_TOKEN = os.getenv("HF_TOKEN", "").strip()
 
-# -------- Data fetch (Open-Meteo + OpenAQ) --------
+# -------- HTTP helper --------
 def _get(url, params=None, headers=None, timeout=12, retries=2, backoff=1.6):
     for i in range(retries + 1):
         try:
@@ -31,6 +25,7 @@ def _get(url, params=None, headers=None, timeout=12, retries=2, backoff=1.6):
                 raise
             time.sleep((backoff ** i) + random.random() * 0.25)
 
+# -------- Geocode (free) --------
 def geocode_city(city:str):
     r = _get("https://nominatim.openstreetmap.org/search",
              params={"q": city, "format": "json", "limit": 1},
@@ -40,6 +35,7 @@ def geocode_city(city:str):
         raise RuntimeError("City not found")
     return {"lat": float(j[0]["lat"]), "lon": float(j[0]["lon"]), "name": j[0]["display_name"]}
 
+# -------- Weather (Open-Meteo, free) --------
 def fetch_open_meteo(lat, lon):
     r = _get("https://api.open-meteo.com/v1/forecast", params={
         "latitude": lat, "longitude": lon,
@@ -49,30 +45,32 @@ def fetch_open_meteo(lat, lon):
     })
     return r.json()
 
-def fetch_openaq_pm25(lat, lon):
-    r = _get("https://api.openaq.org/v3/latest",
-             params={"coordinates": f"{lat},{lon}", "radius": 10000, "limit": 1, "parameter": "pm25"},
-             headers={"User-Agent": "climamind-space"})
-    j = r.json()
-    pm25 = None
-    if j.get("results"):
-        ms = j["results"][0].get("measurements", [])
-        for m in ms:
-            if m.get("parameter") == "pm25":
-                pm25 = m.get("value")
-                break
-    return pm25
+# -------- PM2.5 (Open-Meteo Air-Quality, free; replaces OpenAQ v3 which now needs a key) --------
+def fetch_pm25(lat, lon):
+    try:
+        r = _get("https://air-quality-api.open-meteo.com/v1/air-quality", params={
+            "latitude": lat, "longitude": lon, "hourly": "pm2_5", "timezone": "auto"
+        }, headers={"User-Agent": "climamind-space"})
+        j = r.json()
+        # take the most recent hour
+        hourly = j.get("hourly", {})
+        values = hourly.get("pm2_5") or []
+        if values:
+            return values[-1]
+    except Exception:
+        pass
+    return None  # graceful fallback
 
 def fetch_factors(lat, lon):
     wx = fetch_open_meteo(lat, lon)
-    cur = wx.get("current", {})
+    cur = wx.get("current", {}) or {}
     factors = {
         "temp_c": cur.get("temperature_2m"),
         "rh": cur.get("relative_humidity_2m"),
         "wind_kmh": cur.get("wind_speed_10m"),
         "precip_mm": cur.get("precipitation"),
         "uv": cur.get("uv_index"),
-        "pm25": fetch_openaq_pm25(lat, lon)
+        "pm25": fetch_pm25(lat, lon),
     }
     return {"factors": factors, "raw": wx}
 
@@ -120,7 +118,6 @@ def call_stub(_prompt:str)->str:
     })
 
 def call_hf_model(prompt:str)->str:
-    # Hugging Face Inference API (serverless).
     from huggingface_hub import InferenceClient
     client = InferenceClient(model=MODEL_ID, token=(HF_TOKEN or None))
     out = client.text_generation(
@@ -139,10 +136,9 @@ def _ensure_local_loaded():
     if _local_loaded:
         return
     from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
-    import torch
     bnb_cfg = BitsAndBytesConfig(
         load_in_4bit=True,
-        bnb_4bit_compute_dtype=torch.bfloat16,
+        bnb_4bit_compute_dtype="bfloat16",
         bnb_4bit_use_double_quant=True,
         bnb_4bit_quant_type="nf4",
     )
@@ -158,7 +154,6 @@ def _ensure_local_loaded():
 
 def call_local(prompt:str)->str:
     _ensure_local_loaded()
-    import torch
     if hasattr(tokenizer, "apply_chat_template"):
         messages = [{"role":"user","content":prompt}]
         inputs = tokenizer.apply_chat_template(messages, tokenize=True, return_tensors="pt").to(model.device)
@@ -193,7 +188,6 @@ def reason_answer(loc, coords, factors, query):
     else:
         raw = call_stub(prompt)
 
-    # Extract largest JSON block
     start, end = raw.find("{"), raw.rfind("}")
     if start == -1 or end == -1:
         return {
@@ -247,10 +241,9 @@ demo = gr.Interface(
     title="ClimaMind — K2-Think + Live Climate Data",
     description="Provider = hf_model (Inference API) | local (GPU Space) | stub (offline). Configure env in Space settings.",
     allow_flagging="never",
-    concurrency_limit=2,   # <— replaces deprecated queue(concurrency_count=...)
+    concurrency_limit=2,
 )
 
-# Optional queue for request buffering (no deprecated args)
 demo.queue(max_size=8)
 
 if __name__ == "__main__":