Upload 8 files

FROM python:3.dockerfile

@@ -0,0 +1,21 @@
+FROM python:3.10-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=1 \
+    PORT=7860 \
+    UVICORN_WORKERS=1
+
+WORKDIR /app
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential \
+ && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt /app/requirements.txt
+RUN pip install --upgrade pip && pip install -r requirements.txt
+
+COPY . /app
+RUN mkdir -p /app/data/pdf /app/data/index
+
+CMD ["sh", "-c", "uvicorn app:app --host 0.0.0.0 --port ${PORT}"]

app.py

@@ -1,4 +1,3 @@
-# app.py — FastAPI + Gradio (mounted at "/") for Hugging Face Spaces (Docker SDK)
 from __future__ import annotations
 import os, json, yaml, subprocess, sys, pathlib, traceback
 from typing import List, Dict, Tuple
@@ -6,12 +5,9 @@ from typing import List, Dict, Tuple
 from fastapi import FastAPI, Body
 from fastapi.responses import JSONResponse
 from fastapi.middleware.cors import CORSMiddleware
-
 import gradio as gr
 
-# =========================
-# Config (failsafe loading)
-# =========================
+# ===== config =====
 DEFAULT_CFG = {
     "app_name": "IR/ESG RAG Bot (OpenAI, 8 languages)",
     "embedding_model": "text-embedding-3-large",
@@ -60,39 +56,26 @@ except Exception as e:
     CFG = DEFAULT_CFG
     CFG_ERR = "config.yaml 読み込みエラー: " + str(e)
 
-# =========================
-# Paths / Lazy imports
-# =========================
 INDEX_PATH = pathlib.Path("data/index/faiss.index")
 META_PATH  = pathlib.Path("data/index/meta.jsonl")
 
 def _lazy_imports():
-    """遅延インポート（初期化安定化）"""
     global faiss, np, embed_texts, chat, detect_out_of_scope, sanitize, compliance_block, SCOPE_HINT
-    import faiss
-    import numpy as np
+    import faiss, numpy as np
     from openai_client import embed_texts, chat
     from guardrails import detect_out_of_scope, sanitize, compliance_block, SCOPE_HINT
     return faiss, np, embed_texts, chat, detect_out_of_scope, sanitize, compliance_block, SCOPE_HINT
 
-def _index_exists() -> bool:
-    return INDEX_PATH.exists() and META_PATH.exists()
-
-def _check_api_key() -> bool:
-    return bool(os.getenv("OPENAI_API_KEY"))
+def _index_exists(): return INDEX_PATH.exists() and META_PATH.exists()
+def _check_api_key(): return bool(os.getenv("OPENAI_API_KEY"))
 
-# =========================
-# Retrieval helpers
-# =========================
 _INDEX = None
 _METAS = None
 
 def _ensure_index_loaded():
     global _INDEX, _METAS
-    if _INDEX is not None and _METAS is not None:
-        return
-    if not _index_exists():
-        raise RuntimeError("index_not_ready")
+    if _INDEX is not None and _METAS is not None: return
+    if not _index_exists(): raise RuntimeError("index_not_ready")
     faiss, *_ = _lazy_imports()
     _INDEX = faiss.read_index(str(INDEX_PATH))
     _METAS = [json.loads(l) for l in open(META_PATH, encoding="utf-8")]
@@ -113,16 +96,11 @@ def _search(q: str):
     sims, idxs = sims[0], idxs[0]
     picked, seen = [], set()
     for score, idx in zip(sims, idxs):
-        if score < SCORETH:
-            continue
-        c = _METAS[idx]
-        key = (c["source"], c["page"])
-        if key in seen:
-            continue
-        seen.add(key)
-        picked.append({**c, "score": float(score)})
-        if len(picked) >= TOP_K:
-            break
+        if score < SCORETH: continue
+        c = _METAS[idx]; key = (c["source"], c["page"])
+        if key in seen: continue
+        seen.add(key); picked.append({**c, "score": float(score)})
+        if len(picked) >= TOP_K: break
     return picked
 
 def _format_context(chunks: List[Dict]) -> str:
@@ -143,15 +121,14 @@ _LANG_INSTRUCTIONS = {
     "it": "Rispondi in italiano.",
 }
 
-def generate_answer(q: str, lang: str = "ja") -> Tuple[str, Dict]:
+def generate_answer(q: str, lang: str = "ja"):
     q = (q or "").strip()
-    if not q:
-        return "質問を入力してください。", {}
+    if not q: return "質問を入力してください。", {}
     try:
         _, _, _, chat, detect_out_of_scope, sanitize, compliance_block, SCOPE_HINT = _lazy_imports()
         if detect_out_of_scope(q):
             return f"{SCOPE_HINT}\nIR/ESG関連の事項についてお尋ねください。", {}
-        chunks  = _search(q)
+        chunks = _search(q)
         context = _format_context(chunks)
         lang_note = _LANG_INSTRUCTIONS.get(lang, "Answer in the user's language.")
         user_prompt = (
@@ -161,14 +138,11 @@ def generate_answer(q: str, lang: str = "ja") -> Tuple[str, Dict]:
         )
         messages = [
             {"role": "system", "content": CFG["llm"]["system_prompt"]},
-            {"role": "user",   "content": user_prompt},
+            {"role": "user", "content": user_prompt},
         ]
-        text = chat(
-            messages,
-            model=CFG["llm"]["model"],
-            max_output_tokens=CFG["llm"]["max_output_tokens"],
-            temperature=CFG["llm"]["temperature"],
-        )
+        text = chat(messages, model=CFG["llm"]["model"],
+                    max_output_tokens=CFG["llm"]["max_output_tokens"],
+                    temperature=CFG["llm"]["temperature"])
         text = sanitize(text) + "\n\n" + compliance_block()
         meta = {"citations": [{"source": c["source"], "page": c["page"], "score": round(c["score"], 3)} for c in chunks]}
         return text, meta
@@ -184,34 +158,24 @@ def generate_answer(q: str, lang: str = "ja") -> Tuple[str, Dict]:
 def rebuild_index() -> str:
     if not _check_api_key():
         return "OPENAI_API_KEY が未設定です。コンソール / Secrets に登���してください。"
-    pdf_dir = pathlib.Path("data/pdf")
-    pdf_dir.mkdir(parents=True, exist_ok=True)
-    if not list(pdf_dir.glob("*.pdf")):
-        return "data/pdf/ にPDFがありません。PDFを置いて再実行してください。"
+    pdf_dir = pathlib.Path("data/pdf"); pdf_dir.mkdir(parents=True, exist_ok=True)
+    if not list(pdf_dir.glob("*.pdf")): return "data/pdf/ にPDFがありません。PDFを置いて再実行してください。"
     try:
         out = subprocess.run([sys.executable, "ingest.py"], capture_output=True, text=True, check=True)
-        # キャッシュ破棄
-        global _INDEX, _METAS
-        _INDEX = None
-        _METAS = None
+        global _INDEX, _METAS; _INDEX = None; _METAS = None
         return "✅ インデックス生成完了\n```\n" + (out.stdout[-1200:] or "") + "\n```"
     except subprocess.CalledProcessError as e:
         return f"❌ インデックス生成に失敗\nstdout:\n{e.stdout}\n\nstderr:\n{e.stderr}"
     except Exception as e:
         return "❌ 予期せぬエラー: " + str(e) + "\n" + traceback.format_exc()[-1200:]
 
-# =========================
-# FastAPI (ASGI)
-# =========================
+# ===== FastAPI =====
 app = FastAPI(title=CFG.get("app_name", "RAG Bot"))
 app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"], allow_credentials=False, allow_methods=["*"], allow_headers=["*"],
+    CORSMiddleware, allow_origins=["*"], allow_credentials=False, allow_methods=["*"], allow_headers=["*"]
 )
-
 @app.get("/health")
-def health():
-    return {"status": "ok"}
+def health(): return {"status": "ok"}
 
 @app.post("/api/answer")
 def api_answer(payload: Dict = Body(...)):
@@ -220,19 +184,13 @@ def api_answer(payload: Dict = Body(...)):
 
 @app.post("/api/rebuild")
 def api_rebuild():
-    msg = rebuild_index()
-    return JSONResponse({"message": msg})
-
-# =========================
-# Gradio UI (mounted at "/")
-# =========================
-LANGS  = CFG["languages"]["preferred"]
-LABELS = CFG["languages"].get("labels", {l: l for l in LANGS})
+    return JSONResponse({"message": rebuild_index()})
 
+# ===== Gradio UI mounted at "/" =====
+LANGS = CFG["languages"]["preferred"]
 with gr.Blocks(fill_height=True, title=CFG.get("app_name", "RAG Bot")) as demo:
     gr.Markdown("# IR・ESG開示RAG（OpenAI API）— 8言語対応")
-    if CFG_ERR:
-        gr.Markdown(f"**構成警告**: {CFG_ERR}")
+    if CFG_ERR: gr.Markdown(f"**構成警告**: {CFG_ERR}")
     with gr.Row():
         q = gr.Textbox(label="質問 / Question", lines=3, placeholder="例: 2024年度のGHG排出量(スコープ1-3)は？")
     with gr.Row():
@@ -240,9 +198,7 @@ with gr.Blocks(fill_height=True, title=CFG.get("app_name", "RAG Bot")) as demo:
     with gr.Row():
         ask = gr.Button("回答する / Answer", variant="primary")
         rebuild = gr.Button("インデックス再構築（ingest.py 実行）")
-    ans = gr.Markdown()
-    cites = gr.JSON(label="根拠メタデータ / Citations")
-    log = gr.Markdown()
+    ans = gr.Markdown(); cites = gr.JSON(label="根拠メタデータ / Citations"); log = gr.Markdown()
     ask.click(fn=generate_answer, inputs=[q, lang], outputs=[ans, cites])
     rebuild.click(fn=rebuild_index, outputs=[log])
 

requirements.txt

@@ -1,8 +1,3 @@
-
----
-
-## 3) `requirements.txt`
-```txt
 fastapi==0.112.0
 uvicorn[standard]==0.30.5
 gradio==4.44.1