feat: 安装flash-attn
Browse files- app.py +3 -0
- src/backend/run_eval_suite_harness.py +2 -0
app.py
CHANGED
@@ -20,6 +20,9 @@ from src.display.log_visualizer import log_file_to_html_string
|
|
20 |
from src.display.css_html_js import dark_mode_gradio_js
|
21 |
from src.envs import REFRESH_RATE, REPO_ID, QUEUE_REPO, RESULTS_REPO
|
22 |
from src.logging import setup_logger, log_file
|
|
|
|
|
|
|
23 |
|
24 |
logging.basicConfig(level=logging.INFO)
|
25 |
logger = setup_logger(__name__)
|
|
|
20 |
from src.display.css_html_js import dark_mode_gradio_js
|
21 |
from src.envs import REFRESH_RATE, REPO_ID, QUEUE_REPO, RESULTS_REPO
|
22 |
from src.logging import setup_logger, log_file
|
23 |
+
import subprocess
|
24 |
+
|
25 |
+
subprocess.run("pip install flash-attn --no-build-isolation", env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"}, shell=True)
|
26 |
|
27 |
logging.basicConfig(level=logging.INFO)
|
28 |
logger = setup_logger(__name__)
|
src/backend/run_eval_suite_harness.py
CHANGED
@@ -54,6 +54,8 @@ def run_evaluation(eval_request: EvalRequest, task_names: list, num_fewshot: int
|
|
54 |
write_out=True # Whether to write out an example document and model input, for checking task integrity
|
55 |
)
|
56 |
|
|
|
|
|
57 |
results["config"]["model_dtype"] = eval_request.precision
|
58 |
results["config"]["model_name"] = eval_request.model
|
59 |
results["config"]["model_sha"] = eval_request.revision
|
|
|
54 |
write_out=True # Whether to write out an example document and model input, for checking task integrity
|
55 |
)
|
56 |
|
57 |
+
logger.info(f"Done Tasks: {task_names}")
|
58 |
+
|
59 |
results["config"]["model_dtype"] = eval_request.precision
|
60 |
results["config"]["model_name"] = eval_request.model
|
61 |
results["config"]["model_sha"] = eval_request.revision
|