mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-09-11 20:34:54 +00:00
Trying a few things.
This commit is contained in:
parent
f9abe4cd40
commit
2c12e937ce
2
.github/workflows/build.yaml
vendored
2
.github/workflows/build.yaml
vendored
@ -193,5 +193,7 @@ jobs:
|
||||
export DOCKER_DEVICES=${{ needs.build-and-push.outputs.docker_devices }}
|
||||
export EXTRA_PYTEST="${{ needs.build-and-push.outputs.extra_pytest }}"
|
||||
export HF_TOKEN=${{ secrets.HF_TOKEN }}
|
||||
export ROCR_VISIBLE_DEVICES=0,1
|
||||
echo $DOCKER_IMAGE
|
||||
rocm-smi
|
||||
pytest -s -vv integration-tests ${PYTEST_FLAGS} ${EXTRA_PYTEST}
|
||||
|
@ -534,6 +534,10 @@ def launcher(event_loop):
|
||||
|
||||
yield ContainerLauncherHandle(client, container.name, port)
|
||||
|
||||
import time
|
||||
|
||||
time.sleep(600)
|
||||
|
||||
if not use_flash_attention:
|
||||
del env["USE_FLASH_ATTENTION"]
|
||||
|
||||
@ -589,7 +593,6 @@ def generate_multi():
|
||||
max_new_tokens: int,
|
||||
seed: Optional[int] = None,
|
||||
) -> List[Response]:
|
||||
|
||||
import numpy as np
|
||||
|
||||
arange = np.arange(len(prompts))
|
||||
|
Loading…
Reference in New Issue
Block a user