feat: 安装flash-attn

app.py

@@ -20,6 +20,9 @@ from src.display.log_visualizer import log_file_to_html_string
 from src.display.css_html_js import dark_mode_gradio_js
 from src.envs import REFRESH_RATE, REPO_ID, QUEUE_REPO, RESULTS_REPO
 from src.logging import setup_logger, log_file
+import subprocess
+
+subprocess.run("pip install flash-attn --no-build-isolation", env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"}, shell=True)
 
 logging.basicConfig(level=logging.INFO)
 logger = setup_logger(__name__)

src/backend/run_eval_suite_harness.py

@@ -54,6 +54,8 @@ def run_evaluation(eval_request: EvalRequest, task_names: list, num_fewshot: int
         write_out=True # Whether to write out an example document and model input, for checking task integrity
     )
 
+    logger.info(f"Done Tasks: {task_names}")
+
     results["config"]["model_dtype"] = eval_request.precision
     results["config"]["model_name"] = eval_request.model
     results["config"]["model_sha"] = eval_request.revision