Update streamlit_app.py

streamlit_app.py

@@ -198,9 +198,9 @@ API_KEY_INPUT = settings.text_input("Google API Key (one-time)", value="", type=
 # Default model changed to text-bison@001 (broadly available). Replace if you have another.
 model_input = settings.text_input("Model (short name)", "text-bison@001")
 model_id = model_input.strip() or "text-bison@001"
-# model_arg used with SDK; model_for_url used for HTTP
-model_arg = model_id if not model_id.startswith("models/") else model_id.split("/", 1)[1]
-model_for_url_default = model_arg.split("/", 1)[0] if "@" not in model_arg else model_arg  # keep @ if present
+# model_arg used with SDK; keep as short name like "text-bison@001"
+model_arg = model_id
+model_for_url_default = model_arg
 
 default_prompt = (
     "You are an Indoor Human Behavior Analyzer. Watch the video and produce a detailed, evidence‑based behavioral report focused on human actions, "
@@ -282,19 +282,16 @@ def get_runtime_api_key():
 # ---- Simplified SDK-first + HTTP-fallback layer ----
 def _normalize_model_for_http(model: str) -> str:
     """
-    Return a model identifier appropriate for the HTTP path.
-    Accept inputs like:
-      - "text-bison@001" -> "models/text-bison@001"
-      - "models/text-bison@001" -> "models/text-bison@001"
-      - "models/foo" -> "models/foo"
-      - "foo" -> "models/foo"
+    Return a short model name appropriate for the HTTP path,
+    e.g. "text-bison@001" (no "models/" prefix).
     """
     if not model:
-        return "models/text-bison@001"
+        return "text-bison@001"
     m = model.strip()
+    # If user mistakenly provided "models/..." strip the prefix for HTTP path
     if m.startswith("models/"):
-        return m
-    return f"models/{m}"
+        return m.split("/", 1)[1]
+    return m
 
 def _messages_to_prompt(messages):
     if not messages:
@@ -309,11 +306,10 @@ def _messages_to_prompt(messages):
 def _http_generate(api_key: str, model: str, prompt: str, max_tokens: int):
     host = "https://generativelanguage.googleapis.com"
     norm = _normalize_model_for_http(model)
-    # Try several API paths. norm already contains "models/..."
     candidates = [
-        f"{host}/v1/{norm}:generate",
-        f"{host}/v1beta3/{norm}:generate",
-        f"{host}/v1beta2/{norm}:generate",
+        f"{host}/v1/models/{norm}:generate",
+        f"{host}/v1beta3/models/{norm}:generate",
+        f"{host}/v1beta2/models/{norm}:generate",
     ]
     payload = {"prompt": {"text": prompt}, "maxOutputTokens": int(max_tokens or 512)}
     headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
@@ -326,7 +322,6 @@ def _http_generate(api_key: str, model: str, prompt: str, max_tokens: int):
                     return r.json()
                 except Exception:
                     return {"text": r.text}
-            # Explicit 404 handling: try next candidate
             last_exc = RuntimeError(f"HTTP {r.status_code}: {r.text}")
         except Exception as e:
             last_exc = e
@@ -346,7 +341,6 @@ def responses_generate(model, messages, files, max_output_tokens, api_key):
                     sdk_kwargs["files"] = files
                 return responses_obj.generate(**sdk_kwargs)
         except Exception:
-            # fall through to HTTP fallback
             pass
     # HTTP fallback
     prompt = _messages_to_prompt(messages)
@@ -367,15 +361,11 @@ def call_responses_once(model_used, system_msg, user_msg, fname, max_tokens):
 def extract_text_from_response(response):
     if response is None:
         return None
-    # dict-like (HTTP)
     if isinstance(response, dict):
-        # Try common shapes
-        # 1) "candidates": [ { "content": "...", "text": "..." } ]
         if "candidates" in response and isinstance(response["candidates"], list) and response["candidates"]:
             cand = response["candidates"][0]
             if isinstance(cand, dict):
                 return cand.get("content") or cand.get("text") or response.get("text")
-        # 2) "output": [ { "content": "..."} ] or "output_text"
         if "output" in response and isinstance(response["output"], list):
             pieces = []
             for item in response["output"]:
@@ -389,12 +379,10 @@ def extract_text_from_response(response):
             return response["text"]
         if "outputText" in response and isinstance(response["outputText"], str):
             return response["outputText"]
-        # fallback: try joining string fields
         for k in ("result", "generated_text", "description"):
             if k in response and isinstance(response[k], str):
                 return response[k]
         return None
-    # object-like (SDK)
     try:
         outputs = getattr(response, "output", None) or getattr(response, "candidates", None)
         if outputs:
@@ -414,7 +402,6 @@ def extract_text_from_response(response):
 
 # ---- end compatibility layer ----
 
-# The main generation block (patched, robust)
 if (st.session_state.get("busy") is False) and ('generate_now' in locals() and generate_now):
     if not st.session_state.get("videos"):
         st.error("No video loaded. Use 'Load Video' in the sidebar.")
@@ -485,7 +472,6 @@ if (st.session_state.get("busy") is False) and ('generate_now' in locals() and g
                     model_used = model_arg or "text-bison@001"
                     max_tokens = int(st.session_state.get("max_output_tokens", 1024))
 
-                # Ensure model_used is a short name (SDK accepts it; HTTP will normalize)
                 system_msg = {"role": "system", "content": "You are a helpful assistant that summarizes videos concisely in vivid detail."}
                 user_msg = {"role": "user", "content": prompt_text}
 
@@ -494,7 +480,6 @@ if (st.session_state.get("busy") is False) and ('generate_now' in locals() and g
 
                 out = extract_text_from_response(response)
 
-                # Try to read token info if present
                 meta = getattr(response, "metrics", None) or (response.get("metrics") if isinstance(response, dict) else None) or {}
                 output_tokens = 0
                 try:
@@ -505,7 +490,6 @@ if (st.session_state.get("busy") is False) and ('generate_now' in locals() and g
                 except Exception:
                     output_tokens = 0
 
-                # Retry strategies if no output
                 if (not out or output_tokens == 0) and model_used:
                     retry_prompt = "Summarize the video content briefly and vividly (2-4 paragraphs)."
                     try:
@@ -530,7 +514,6 @@ if (st.session_state.get("busy") is False) and ('generate_now' in locals() and g
                 st.subheader("Analysis Result")
                 st.markdown(out or "_(no text returned)_")
 
-                # Clean up compressed temporary file only (don't remove original unless it was a true temp)
                 try:
                     if reupload_needed:
                         try:
@@ -540,8 +523,6 @@ if (st.session_state.get("busy") is False) and ('generate_now' in locals() and g
                                     p.unlink(missing_ok=True)
                         except Exception:
                             pass
-                        # don't delete original unless it's intended to be removed (conservative)
-                        # If you want original removed, enable a setting and handle carefully.
                 except Exception:
                     pass
 
@@ -590,7 +571,8 @@ with st.sidebar.expander("Manage uploads", expanded=False):
         st.session_state["uploaded_file"] = None
         st.session_state["processed_file"] = None
         st.session_state["last_loaded_path"] = ""
-        st.session_state[""] = None
+        st.session_state["analysis_out"] = ""
+        st.session_state["file_hash"] = None
         try:
             fname = file_name_or_id(st.session_state.get("uploaded_file"))
             if fname and delete_file and HAS_GENAI: