feat: csv report

README.md

@@ -13,5 +13,5 @@ pinned: false
 
 ```shell
 uv sync
-uv run python app.py
+uv run app.py
 ```

app.py

@@ -1,3 +1,4 @@
+import csv
 import json
 import os
 import tempfile
@@ -21,6 +22,7 @@ def process_jsonl_file(jsonl_file_path: str, api_key: str):
         file_name_with_ext = os.path.basename(jsonl_file_path)
         file_name, _ = os.path.splitext(file_name_with_ext)
 
+        output_file = None
         with tempfile.NamedTemporaryFile(
             delete=False,
             prefix=f"{file_name}-report-",
@@ -30,10 +32,25 @@ def process_jsonl_file(jsonl_file_path: str, api_key: str):
         ) as temp_file:
             temp_file.write(html_content)
             output_file = temp_file.name
-        return output_file, ""
+        
+        output_csv = None
+        keys = json_data[0].keys()
+        with tempfile.NamedTemporaryFile(
+            delete=False,
+            prefix=f"{file_name}-report-",
+            suffix=".csv",
+            mode="w",
+            encoding="utf-8",
+        ) as temp_file:
+            dict_writer = csv.DictWriter(temp_file, fieldnames=keys)
+            dict_writer.writeheader()
+            dict_writer.writerows(json_data)
+            output_csv = temp_file.name
+
+        return output_file, output_csv, ""
 
     except Exception as e:
-        return None, e
+        return None, None, e
 
 
 # Gradioデモ
@@ -45,13 +62,14 @@ with gr.Blocks() as reporting:
     gr.Markdown("APIキーの発行は[こちら](https://aistudio.google.com/app/apikey)")
     process_button = gr.Button("レポートを作成")
 
-    output_file = gr.File(label="セルフ評価レポート")
+    output_file = gr.File(label="セルフ評価レポート(HTML)")
+    output_csv = gr.File(label="セルフ評価レポート(CSV)")
     output_text = gr.Textbox(label="システムメッセージ")
 
     process_button.click(
         process_jsonl_file,
         inputs=[jsonl_input, api_key_input],
-        outputs=[output_file, output_text],
+        outputs=[output_file, output_csv, output_text],
     )
 
 llm_jp_3 = "llm-jp/llm-jp-3-1.8b"

utils.py

@@ -79,15 +79,16 @@ def evaluate(results: list[dict], api_key: str, batch_size: int = 10) -> list[di
             )
 
         for result, score in zip(batch_results, scores):
-            evaluations.append(
-                {
-                    "input": result["input"],
-                    "output": result["output"],
-                    "eval_aspect": result.get("eval_aspect"),
-                    "target": result.get("target"),
-                    "score": score,
-                }
-            )
+            evaluation = {
+                "input": result["input"],
+                "output": result["output"],
+                "score": score,
+            }
+            if "eval_aspect" in result:
+                evaluation["eval_aspect"] = result["eval_aspect"]
+            if "target" in result:
+                evaluation["target"] = result["target"]
+            evaluations.append(evaluation)
 
     return evaluations