Spaces:
Sleeping
Sleeping
fast api added
Browse files- app.py +89 -0
- requirements.txt +4 -0
app.py
CHANGED
|
@@ -10,6 +10,9 @@ from pathlib import Path
|
|
| 10 |
from typing import List, Dict, Any, Optional
|
| 11 |
|
| 12 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
# ZeroGPU対応: spacesパッケージをインポート(デコレータ用)
|
| 15 |
try:
|
|
@@ -41,6 +44,18 @@ status_lock = threading.Lock()
|
|
| 41 |
MODEL_TYPE = os.getenv("MODEL_TYPE", "transformers")
|
| 42 |
HF_MODEL_REPO = os.getenv("HF_MODEL_REPO", "meta-llama/Llama-3.2-3B-Instruct")
|
| 43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
|
| 45 |
def _set_status(message: str) -> None:
|
| 46 |
"""ステータスメッセージを更新"""
|
|
@@ -214,6 +229,76 @@ with gr.Blocks(title="LLMView Multi-Model", theme=gr.themes.Soft()) as demo:
|
|
| 214 |
)
|
| 215 |
|
| 216 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
# ZeroGPU対応: 起動時に検出されるように、デコレータ付き関数を定義
|
| 218 |
@spaces.GPU
|
| 219 |
def _gpu_init_function():
|
|
@@ -221,6 +306,10 @@ def _gpu_init_function():
|
|
| 221 |
pass
|
| 222 |
|
| 223 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 224 |
if __name__ == "__main__":
|
| 225 |
# Hugging Face Spaces用の設定
|
| 226 |
# GPU要求を確実に検出させる
|
|
|
|
| 10 |
from typing import List, Dict, Any, Optional
|
| 11 |
|
| 12 |
import gradio as gr
|
| 13 |
+
from fastapi import FastAPI, HTTPException
|
| 14 |
+
from fastapi.responses import JSONResponse
|
| 15 |
+
from pydantic import BaseModel, Field
|
| 16 |
|
| 17 |
# ZeroGPU対応: spacesパッケージをインポート(デコレータ用)
|
| 18 |
try:
|
|
|
|
| 44 |
MODEL_TYPE = os.getenv("MODEL_TYPE", "transformers")
|
| 45 |
HF_MODEL_REPO = os.getenv("HF_MODEL_REPO", "meta-llama/Llama-3.2-3B-Instruct")
|
| 46 |
|
| 47 |
+
# FastAPI用のリクエスト/レスポンスモデル
|
| 48 |
+
class WordTreeRequest(BaseModel):
|
| 49 |
+
prompt_text: str = Field(..., description="生成に使用するプロンプト")
|
| 50 |
+
root_text: str = Field("", description="任意のルートテキスト")
|
| 51 |
+
top_k: int = Field(5, ge=1, le=50, description="取得する候補数")
|
| 52 |
+
max_depth: int = Field(10, ge=1, le=50, description="探索深さ")
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
class WordTreeResponse(BaseModel):
|
| 56 |
+
text: str
|
| 57 |
+
probability: float
|
| 58 |
+
|
| 59 |
|
| 60 |
def _set_status(message: str) -> None:
|
| 61 |
"""ステータスメッセージを更新"""
|
|
|
|
| 229 |
)
|
| 230 |
|
| 231 |
|
| 232 |
+
# FastAPIアプリを作成(外部APIアクセス用)
|
| 233 |
+
fastapi_app = FastAPI(
|
| 234 |
+
title="LLMView Multi-Model API",
|
| 235 |
+
description="LLMView の単語ツリー構築 API。/build_word_tree にPOSTしてください。",
|
| 236 |
+
version="1.0.0",
|
| 237 |
+
)
|
| 238 |
+
|
| 239 |
+
|
| 240 |
+
@fastapi_app.get("/")
|
| 241 |
+
def root() -> Dict[str, str]:
|
| 242 |
+
"""簡易案内"""
|
| 243 |
+
return {
|
| 244 |
+
"message": "LLMView Multi-Model API",
|
| 245 |
+
"status_endpoint": "/health",
|
| 246 |
+
"build_endpoint": "/build_word_tree",
|
| 247 |
+
"gradio_ui": "/",
|
| 248 |
+
}
|
| 249 |
+
|
| 250 |
+
|
| 251 |
+
@fastapi_app.get("/health")
|
| 252 |
+
def health() -> Dict[str, Any]:
|
| 253 |
+
"""状態確認"""
|
| 254 |
+
with status_lock:
|
| 255 |
+
current_status = status_message
|
| 256 |
+
|
| 257 |
+
return {
|
| 258 |
+
"model_loaded": adapter is not None,
|
| 259 |
+
"status": current_status,
|
| 260 |
+
"model_type": MODEL_TYPE,
|
| 261 |
+
"model_path": HF_MODEL_REPO if MODEL_TYPE == "transformers" else None,
|
| 262 |
+
}
|
| 263 |
+
|
| 264 |
+
|
| 265 |
+
@spaces.GPU # ZeroGPU対応
|
| 266 |
+
@fastapi_app.post("/build_word_tree", response_model=List[WordTreeResponse])
|
| 267 |
+
def api_build_word_tree(payload: WordTreeRequest) -> List[WordTreeResponse]:
|
| 268 |
+
"""単語ツリーを構築(FastAPIエンドポイント)"""
|
| 269 |
+
if not payload.prompt_text.strip():
|
| 270 |
+
raise HTTPException(status_code=400, detail="prompt_text を入力してください。")
|
| 271 |
+
|
| 272 |
+
if adapter is None:
|
| 273 |
+
with status_lock:
|
| 274 |
+
current_status = status_message
|
| 275 |
+
raise HTTPException(
|
| 276 |
+
status_code=503, detail=f"モデル準備中です: {current_status}"
|
| 277 |
+
)
|
| 278 |
+
|
| 279 |
+
try:
|
| 280 |
+
results = adapter.build_word_tree(
|
| 281 |
+
prompt_text=payload.prompt_text,
|
| 282 |
+
root_text=payload.root_text,
|
| 283 |
+
top_k=payload.top_k,
|
| 284 |
+
max_depth=payload.max_depth,
|
| 285 |
+
)
|
| 286 |
+
|
| 287 |
+
if not results:
|
| 288 |
+
# ダミー結果を返す
|
| 289 |
+
results = [
|
| 290 |
+
{"text": "候補が生成されませんでした", "probability": 0.0}
|
| 291 |
+
]
|
| 292 |
+
|
| 293 |
+
return [WordTreeResponse(**item) for item in results]
|
| 294 |
+
except Exception as exc:
|
| 295 |
+
import traceback
|
| 296 |
+
traceback.print_exc()
|
| 297 |
+
raise HTTPException(
|
| 298 |
+
status_code=500, detail=f"エラーが発生しました: {exc}"
|
| 299 |
+
)
|
| 300 |
+
|
| 301 |
+
|
| 302 |
# ZeroGPU対応: 起動時に検出されるように、デコレータ付き関数を定義
|
| 303 |
@spaces.GPU
|
| 304 |
def _gpu_init_function():
|
|
|
|
| 306 |
pass
|
| 307 |
|
| 308 |
|
| 309 |
+
# GradioアプリにFastAPIを統合
|
| 310 |
+
demo.fastapi_app = fastapi_app
|
| 311 |
+
|
| 312 |
+
|
| 313 |
if __name__ == "__main__":
|
| 314 |
# Hugging Face Spaces用の設定
|
| 315 |
# GPU要求を確実に検出させる
|
requirements.txt
CHANGED
|
@@ -6,6 +6,10 @@
|
|
| 6 |
# Gradio - Web UI(最新版でセキュリティ脆弱性を修正)
|
| 7 |
gradio>=4.43.0
|
| 8 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
# Hugging Face Spaces
|
| 10 |
spaces
|
| 11 |
|
|
|
|
| 6 |
# Gradio - Web UI(最新版でセキュリティ脆弱性を修正)
|
| 7 |
gradio>=4.43.0
|
| 8 |
|
| 9 |
+
# FastAPI - 外部APIアクセス用
|
| 10 |
+
fastapi>=0.111.0
|
| 11 |
+
pydantic>=2.0.0
|
| 12 |
+
|
| 13 |
# Hugging Face Spaces
|
| 14 |
spaces
|
| 15 |
|