Update src/streamlit_app.py

src/streamlit_app.py

@@ -2,23 +2,11 @@ import os
 import time
 import threading
 import requests
-import io
 
 import streamlit as st
 import torch
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, TextIteratorStreamer
 
-# Optional parsing libs (safe fallbacks)
-try:
-    from pypdf import PdfReader  # pip install pypdf
-except Exception:
-    PdfReader = None
-
-try:
-    import pandas as pd  # pip install pandas
-except Exception:
-    pd = None
-
 # Optional LangSmith
 try:
     from langsmith import Client as LangSmithClient
@@ -103,75 +91,13 @@ with st.sidebar:
         height=160,
     )
 
-    local_context_text = st.text_area(
+    local_context = st.text_area(
         "Local context (optional)",
-        height=120,
-        placeholder="Extra context to append after web snippets…",
-    )
-
-    uploaded_files = st.file_uploader(
-        "Upload context files",
-        accept_multiple_files=True,
-        type=None,
-        help="PDF, TXT, CSV, MD, JSON, etc.",
+        height=140,
+        placeholder="Extra context appended after web snippets…",
     )
 
 
-# =========================
-# FILE PARSING
-# =========================
-def safe_decode(b: bytes) -> str:
-    for enc in ("utf-8", "utf-16", "latin-1"):
-        try:
-            return b.decode(enc)
-        except Exception:
-            pass
-    return b.decode("utf-8", errors="ignore")
-
-
-def parse_file_to_text(file) -> str:
-    name = (file.name or "").lower()
-    raw = file.getvalue()
-
-    if name.endswith(".pdf") and PdfReader:
-        try:
-            reader = PdfReader(io.BytesIO(raw))
-            parts = []
-            for p in reader.pages:
-                t = (p.extract_text() or "").strip()
-                if t:
-                    parts.append(t)
-            return "\n\n".join(parts).strip()
-        except Exception as e:
-            return f"[PDF parse error: {e}]"
-
-    if name.endswith(".csv") and pd:
-        try:
-            df = pd.read_csv(io.BytesIO(raw))
-            return df.to_csv(index=False)
-        except Exception:
-            return safe_decode(raw)
-
-    return safe_decode(raw).strip()
-
-
-def build_local_context(text_block: str, files) -> str:
-    chunks = []
-    if text_block and text_block.strip():
-        chunks.append(text_block.strip())
-
-    if files:
-        for f in files:
-            parsed = parse_file_to_text(f)
-            if parsed and parsed.strip():
-                chunks.append(f"\n\n--- {f.name} ---\n{parsed.strip()}")
-
-    return "\n\n".join(chunks).strip()
-
-
-local_context = build_local_context(local_context_text, uploaded_files)
-
-
 # =========================
 # WEB SEARCH (ALWAYS ON)
 # =========================
@@ -198,7 +124,8 @@ def web_search_snippets(query: str):
 
     snippets = []
     for item in data.get("web", {}).get("results", [])[:TOP_K_SEARCH]:
-        desc = (item.get("description") or "").replace("<strong>", "").replace("</strong>", "").strip()
+        desc = (item.get("description") or "")
+        desc = desc.replace("<strong>", "").replace("</strong>", "").strip()
         if desc:
             snippets.append(desc)
 
@@ -214,6 +141,7 @@ def truncate_context(web_ctx: str, local_ctx: str, system: str, question: str) -
     base = f"\n{system}\n{question}\n"
     base_tokens = tokenizer.encode(base)
     budget = MAX_INPUT_TOKENS - len(base_tokens)
+
     if budget <= 0:
         return ""
 
@@ -267,7 +195,6 @@ def handle_feedback(idx: int):
                 run_id=msg["run_id"],
                 key="thumb_rating",
                 score=score,
-                comment="thumbs_up" if score else "thumbs_down",
             )
         except Exception:
             pass
@@ -281,19 +208,12 @@ for i, msg in enumerate(st.session_state.messages):
         st.markdown(msg["content"])
 
         if msg["role"] == "assistant":
-            # Light, normal-looking stats
             st.caption(
-                f"{msg['search_time']:.2f}s search • {msg['gen_time']:.2f}s gen • "
+                f"{msg['search_time']:.2f}s • {msg['gen_time']:.2f}s • "
                 f"{msg['tps']:.1f} tok/s • in {msg['input_tokens']} • out {msg['output_tokens']}"
             )
 
-            # Small inspector (collapsed)
             with st.expander("Inspect context"):
-                st.markdown("**System**")
-                st.code(msg.get("system_prompt", ""), language="text")
-                st.markdown("**Question**")
-                st.code(msg.get("question", ""), language="text")
-                st.markdown("**Prompt (sent to model)**")
                 st.code(msg.get("prompt", ""), language="text")
 
             key = f"fb_{i}"
@@ -323,35 +243,30 @@ if query:
 if st.session_state.messages and st.session_state.messages[-1]["role"] == "user":
     question = st.session_state.messages[-1]["content"]
 
-    # web search
     web_ctx, search_time = web_search_snippets(question)
 
-    # truncate final context
-    final_context = truncate_context(web_ctx, local_context, system_prompt, question)
+    final_context = truncate_context(
+        web_ctx,
+        local_context,
+        system_prompt,
+        question,
+    )
 
-    # prompt sent to model
     prompt = f"{final_context}\n{system_prompt}\n{question}\n"
-
     input_tokens = count_tokens(prompt)
 
-    # LangSmith run (optional)
     run_id = None
     if ls_client:
         try:
             run = ls_client.create_run(
                 name="teapot_chat",
                 run_type="llm",
-                inputs={
-                    "system_prompt": system_prompt,
-                    "question": question,
-                    "prompt": prompt,
-                },
+                inputs={"prompt": prompt, "question": question},
             )
             run_id = run.id
         except Exception:
             pass
 
-    # stream normally in chat
     with st.chat_message("assistant"):
         placeholder = st.empty()
         start = time.perf_counter()
@@ -366,16 +281,11 @@ if st.session_state.messages and st.session_state.messages[-1]["role"] == "user"
         tps = output_tokens / gen_time if gen_time > 0 else 0.0
 
         st.caption(
-            f"{search_time:.2f}s search • {gen_time:.2f}s gen • "
+            f"{search_time:.2f}s • {gen_time:.2f}s • "
             f"{tps:.1f} tok/s • in {input_tokens} • out {output_tokens}"
         )
 
         with st.expander("Inspect context"):
-            st.markdown("**System**")
-            st.code(system_prompt, language="text")
-            st.markdown("**Question**")
-            st.code(question, language="text")
-            st.markdown("**Prompt (sent to model)**")
             st.code(prompt, language="text")
 
     if ls_client and run_id:
@@ -388,8 +298,6 @@ if st.session_state.messages and st.session_state.messages[-1]["role"] == "user"
         {
             "role": "assistant",
             "content": final_text,
-            "system_prompt": system_prompt,
-            "question": question,
             "prompt": prompt,
             "search_time": search_time,
             "gen_time": gen_time,