Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,8 +2,8 @@ import os
|
|
| 2 |
import io
|
| 3 |
import json
|
| 4 |
import hashlib
|
| 5 |
-
import pathlib
|
| 6 |
import gradio as gr
|
|
|
|
| 7 |
|
| 8 |
from pipelines.openai_ingest import (
|
| 9 |
extract_text_with_openai,
|
|
@@ -20,21 +20,33 @@ from pipelines.utils import detect_filetype, load_doc_text
|
|
| 20 |
|
| 21 |
APP_TITLE = "候補者インテーク & レジュメ標準化(OpenAI版)"
|
| 22 |
|
|
|
|
| 23 |
|
| 24 |
-
def
|
| 25 |
-
"""gr.Files v4 用の堅牢リーダ。
|
| 26 |
-
- type="filepath" の場合: f は str/Path
|
| 27 |
-
- type="binary" の場合: f は UploadedFile ライク(.name/.read())
|
| 28 |
-
戻り値: (filename, bytes)
|
| 29 |
"""
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
|
|
|
|
| 36 |
|
| 37 |
-
def process_resumes(files, candidate_id: str, additional_notes: str = ""):
|
| 38 |
if not files:
|
| 39 |
raise gr.Error("少なくとも1ファイルをアップロードしてください。")
|
| 40 |
|
|
@@ -42,29 +54,20 @@ def process_resumes(files, candidate_id: str, additional_notes: str = ""):
|
|
| 42 |
raw_texts = []
|
| 43 |
|
| 44 |
for f in files:
|
| 45 |
-
|
| 46 |
filetype = detect_filetype(fname, raw_bytes)
|
| 47 |
|
| 48 |
# 1) テキスト抽出:画像/PDFはOpenAI Vision OCR、docx/txtは生文面+OpenAI整形
|
| 49 |
if filetype in {"pdf", "image"}:
|
| 50 |
text = extract_text_with_openai(raw_bytes, filename=fname, filetype=filetype)
|
| 51 |
-
elif filetype in {"docx", "txt"}:
|
| 52 |
-
base_text = load_doc_text(filetype, raw_bytes)
|
| 53 |
-
# 生テキストをOpenAIへ渡し、整形した全文を返す
|
| 54 |
-
text = extract_text_with_openai(base_text.encode("utf-8"), filename=fname, filetype="txt")
|
| 55 |
else:
|
| 56 |
-
|
| 57 |
-
try:
|
| 58 |
-
base_text = raw_bytes.decode("utf-8", errors="ignore")
|
| 59 |
-
except Exception:
|
| 60 |
-
base_text = ""
|
| 61 |
text = extract_text_with_openai(base_text.encode("utf-8"), filename=fname, filetype="txt")
|
| 62 |
|
| 63 |
raw_texts.append({"filename": fname, "text": text})
|
| 64 |
|
| 65 |
-
# 2) OpenAIでセクション構造化
|
| 66 |
structured = structure_with_openai(text)
|
| 67 |
-
# 念のためルールベース正規化も適用(期間抽出など補助)
|
| 68 |
normalized = normalize_resume({
|
| 69 |
"work_experience": structured.get("work_experience_raw", ""),
|
| 70 |
"education": structured.get("education_raw", ""),
|
|
@@ -78,10 +81,10 @@ def process_resumes(files, candidate_id: str, additional_notes: str = ""):
|
|
| 78 |
"normalized": normalized,
|
| 79 |
})
|
| 80 |
|
| 81 |
-
# 3)
|
| 82 |
merged = merge_normalized_records([r["normalized"] for r in partial_records])
|
| 83 |
|
| 84 |
-
# 4)
|
| 85 |
merged_text = "\n\n".join([r["text"] for r in partial_records])
|
| 86 |
skills = extract_skills(merged_text, {
|
| 87 |
"work_experience": merged.get("raw_sections", {}).get("work_experience", ""),
|
|
@@ -97,14 +100,13 @@ def process_resumes(files, candidate_id: str, additional_notes: str = ""):
|
|
| 97 |
# 6) 品質スコア
|
| 98 |
score = compute_quality_score(merged_text, merged)
|
| 99 |
|
| 100 |
-
# 7)
|
| 101 |
summaries = summarize_with_openai(merged_text)
|
| 102 |
|
| 103 |
# 8) 構造化出力
|
| 104 |
-
candidate_id_final = candidate_id or hashlib.sha256(merged_text.encode("utf-8")).hexdigest()[:16]
|
| 105 |
result_json = {
|
| 106 |
-
"candidate_id":
|
| 107 |
-
"files": [
|
| 108 |
"merged": merged,
|
| 109 |
"skills": skills,
|
| 110 |
"quality_score": score,
|
|
@@ -127,22 +129,24 @@ def process_resumes(files, candidate_id: str, additional_notes: str = ""):
|
|
| 127 |
pdf_path=f"candidates/{file_hash}.anon.pdf",
|
| 128 |
)
|
| 129 |
|
| 130 |
-
anon_pdf = (
|
| 131 |
|
| 132 |
-
#
|
| 133 |
return (
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
summaries.get("300chars", ""),
|
| 138 |
summaries.get("100chars", ""),
|
| 139 |
summaries.get("onesent", ""),
|
| 140 |
anon_pdf,
|
| 141 |
-
|
| 142 |
)
|
| 143 |
|
| 144 |
|
| 145 |
-
|
|
|
|
|
|
|
| 146 |
gr.Markdown(f"# {APP_TITLE}\n複数ファイルを統合→OpenAIで読み込み/構造化/要約→匿名化→Datasets保存")
|
| 147 |
|
| 148 |
with gr.Row():
|
|
@@ -150,7 +154,7 @@ with gr.Blocks(title=APP_TITLE) as demo:
|
|
| 150 |
label="レジュメ類 (PDF/画像/Word/テキスト) 複数可",
|
| 151 |
file_count="multiple",
|
| 152 |
file_types=[".pdf", ".png", ".jpg", ".jpeg", ".tiff", ".bmp", ".docx", ".txt"],
|
| 153 |
-
type="filepath", #
|
| 154 |
)
|
| 155 |
candidate_id = gr.Textbox(label="候補者ID(任意。未入力なら自動生成)")
|
| 156 |
notes = gr.Textbox(label="補足メモ(任意)", lines=3)
|
|
@@ -161,11 +165,11 @@ with gr.Blocks(title=APP_TITLE) as demo:
|
|
| 161 |
out_json = gr.Code(label="統合出力 (JSON)")
|
| 162 |
|
| 163 |
with gr.Tab("抽出スキル"):
|
| 164 |
-
# gr.JSON は
|
| 165 |
-
out_skills = gr.Code(label="スキル一覧(JSON)")
|
| 166 |
|
| 167 |
with gr.Tab("品質スコア"):
|
| 168 |
-
out_score = gr.Code(label="品質評価(JSON)")
|
| 169 |
|
| 170 |
with gr.Tab("要約 (300/100/1文)"):
|
| 171 |
out_sum_300 = gr.Textbox(label="300字要約")
|
|
@@ -176,7 +180,7 @@ with gr.Blocks(title=APP_TITLE) as demo:
|
|
| 176 |
out_pdf = gr.File(label="匿名PDFダウンロード")
|
| 177 |
|
| 178 |
with gr.Tab("Datasets 保存ログ"):
|
| 179 |
-
out_commit = gr.Code(label="コミット情報")
|
| 180 |
|
| 181 |
run_btn.click(
|
| 182 |
process_resumes,
|
|
@@ -186,9 +190,7 @@ with gr.Blocks(title=APP_TITLE) as demo:
|
|
| 186 |
|
| 187 |
|
| 188 |
if __name__ == "__main__":
|
| 189 |
-
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
# 実行環境が localhost にアクセスできない場合のフォールバック
|
| 194 |
-
demo.launch(share=True)
|
|
|
|
| 2 |
import io
|
| 3 |
import json
|
| 4 |
import hashlib
|
|
|
|
| 5 |
import gradio as gr
|
| 6 |
+
from typing import Tuple, List, Union
|
| 7 |
|
| 8 |
from pipelines.openai_ingest import (
|
| 9 |
extract_text_with_openai,
|
|
|
|
| 20 |
|
| 21 |
APP_TITLE = "候補者インテーク & レジュメ標準化(OpenAI版)"
|
| 22 |
|
| 23 |
+
# ---- helpers ---------------------------------------------------------------
|
| 24 |
|
| 25 |
+
def _read_file_input(item: Union[str, "gradio.files.TempFile"]) -> Tuple[bytes, str]:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
"""
|
| 27 |
+
Gradio v4.44 の Files(type='filepath') は str パスを返す。
|
| 28 |
+
互換のため、パス/ファイルライク双方を許容して (bytes, filename) を返す。
|
| 29 |
+
"""
|
| 30 |
+
if isinstance(item, str):
|
| 31 |
+
with open(item, "rb") as rf:
|
| 32 |
+
data = rf.read()
|
| 33 |
+
name = os.path.basename(item)
|
| 34 |
+
return data, name
|
| 35 |
+
# UploadedFile 等(念のため)
|
| 36 |
+
if hasattr(item, "read"):
|
| 37 |
+
data = item.read()
|
| 38 |
+
name = getattr(item, "name", "uploaded")
|
| 39 |
+
return data, os.path.basename(name)
|
| 40 |
+
raise ValueError("Unsupported file input type")
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
def _as_json_code(obj) -> str:
|
| 44 |
+
return json.dumps(obj, ensure_ascii=False, indent=2)
|
| 45 |
+
|
| 46 |
|
| 47 |
+
# ---- core pipeline ---------------------------------------------------------
|
| 48 |
|
| 49 |
+
def process_resumes(files: List[Union[str, "gradio.files.TempFile"]], candidate_id: str, additional_notes: str = ""):
|
| 50 |
if not files:
|
| 51 |
raise gr.Error("少なくとも1ファイルをアップロードしてください。")
|
| 52 |
|
|
|
|
| 54 |
raw_texts = []
|
| 55 |
|
| 56 |
for f in files:
|
| 57 |
+
raw_bytes, fname = _read_file_input(f)
|
| 58 |
filetype = detect_filetype(fname, raw_bytes)
|
| 59 |
|
| 60 |
# 1) テキスト抽出:画像/PDFはOpenAI Vision OCR、docx/txtは生文面+OpenAI整形
|
| 61 |
if filetype in {"pdf", "image"}:
|
| 62 |
text = extract_text_with_openai(raw_bytes, filename=fname, filetype=filetype)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
else:
|
| 64 |
+
base_text = load_doc_text(filetype, raw_bytes)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
text = extract_text_with_openai(base_text.encode("utf-8"), filename=fname, filetype="txt")
|
| 66 |
|
| 67 |
raw_texts.append({"filename": fname, "text": text})
|
| 68 |
|
| 69 |
+
# 2) OpenAIでセクション構造化 → ルール整形
|
| 70 |
structured = structure_with_openai(text)
|
|
|
|
| 71 |
normalized = normalize_resume({
|
| 72 |
"work_experience": structured.get("work_experience_raw", ""),
|
| 73 |
"education": structured.get("education_raw", ""),
|
|
|
|
| 81 |
"normalized": normalized,
|
| 82 |
})
|
| 83 |
|
| 84 |
+
# 3) 統合
|
| 85 |
merged = merge_normalized_records([r["normalized"] for r in partial_records])
|
| 86 |
|
| 87 |
+
# 4) スキル抽出
|
| 88 |
merged_text = "\n\n".join([r["text"] for r in partial_records])
|
| 89 |
skills = extract_skills(merged_text, {
|
| 90 |
"work_experience": merged.get("raw_sections", {}).get("work_experience", ""),
|
|
|
|
| 100 |
# 6) 品質スコア
|
| 101 |
score = compute_quality_score(merged_text, merged)
|
| 102 |
|
| 103 |
+
# 7) 要約
|
| 104 |
summaries = summarize_with_openai(merged_text)
|
| 105 |
|
| 106 |
# 8) 構造化出力
|
|
|
|
| 107 |
result_json = {
|
| 108 |
+
"candidate_id": candidate_id or hashlib.sha256(merged_text.encode("utf-8")).hexdigest()[:16],
|
| 109 |
+
"files": [os.path.basename(_read_file_input(f)[1]) if isinstance(f, str) else getattr(f, "name", "uploaded") for f in files],
|
| 110 |
"merged": merged,
|
| 111 |
"skills": skills,
|
| 112 |
"quality_score": score,
|
|
|
|
| 129 |
pdf_path=f"candidates/{file_hash}.anon.pdf",
|
| 130 |
)
|
| 131 |
|
| 132 |
+
anon_pdf = (result_json["candidate_id"] + ".anon.pdf", anon_pdf_bytes)
|
| 133 |
|
| 134 |
+
# 返却はすべて文字列 or ファイル
|
| 135 |
return (
|
| 136 |
+
_as_json_code(result_json),
|
| 137 |
+
_as_json_code(skills),
|
| 138 |
+
_as_json_code(score),
|
| 139 |
summaries.get("300chars", ""),
|
| 140 |
summaries.get("100chars", ""),
|
| 141 |
summaries.get("onesent", ""),
|
| 142 |
anon_pdf,
|
| 143 |
+
_as_json_code(commit_info or {"status": "skipped (DATASET_REPO not set)"}),
|
| 144 |
)
|
| 145 |
|
| 146 |
|
| 147 |
+
# ---- UI --------------------------------------------------------------------
|
| 148 |
+
|
| 149 |
+
with gr.Blocks(title=APP_TITLE, analytics_enabled=False) as demo:
|
| 150 |
gr.Markdown(f"# {APP_TITLE}\n複数ファイルを統合→OpenAIで読み込み/構造化/要約→匿名化→Datasets保存")
|
| 151 |
|
| 152 |
with gr.Row():
|
|
|
|
| 154 |
label="レジュメ類 (PDF/画像/Word/テキスト) 複数可",
|
| 155 |
file_count="multiple",
|
| 156 |
file_types=[".pdf", ".png", ".jpg", ".jpeg", ".tiff", ".bmp", ".docx", ".txt"],
|
| 157 |
+
type="filepath", # ← 重要: 'file' はNG
|
| 158 |
)
|
| 159 |
candidate_id = gr.Textbox(label="候補者ID(任意。未入力なら自動生成)")
|
| 160 |
notes = gr.Textbox(label="補足メモ(任意)", lines=3)
|
|
|
|
| 165 |
out_json = gr.Code(label="統合出力 (JSON)")
|
| 166 |
|
| 167 |
with gr.Tab("抽出スキル"):
|
| 168 |
+
# gr.JSON は 4.44 でスキーマ事故発生 → Code に置換
|
| 169 |
+
out_skills = gr.Code(label="スキル一覧 (JSON)")
|
| 170 |
|
| 171 |
with gr.Tab("品質スコア"):
|
| 172 |
+
out_score = gr.Code(label="品質評価 (JSON)")
|
| 173 |
|
| 174 |
with gr.Tab("要約 (300/100/1文)"):
|
| 175 |
out_sum_300 = gr.Textbox(label="300字要約")
|
|
|
|
| 180 |
out_pdf = gr.File(label="匿名PDFダウンロード")
|
| 181 |
|
| 182 |
with gr.Tab("Datasets 保存ログ"):
|
| 183 |
+
out_commit = gr.Code(label="コミット情報 (JSON)")
|
| 184 |
|
| 185 |
run_btn.click(
|
| 186 |
process_resumes,
|
|
|
|
| 190 |
|
| 191 |
|
| 192 |
if __name__ == "__main__":
|
| 193 |
+
# ローカル不可な環境では share=True を強制(環境変数で上書き可)
|
| 194 |
+
share_env = os.environ.get("GRADIO_SHARE", "true").lower()
|
| 195 |
+
share_flag = share_env in ("1", "true", "yes", "on")
|
| 196 |
+
demo.launch(server_name="0.0.0.0", server_port=7860, share=share_flag)
|
|
|
|
|
|