-
Notifications
You must be signed in to change notification settings - Fork 531
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Mount cached mode #4369
base: master
Are you sure you want to change the base?
Mount cached mode #4369
Conversation
* fix permissions * fix permissions
* Add serve for gemma and fix mixtral dependency * Add hf token * fix model len * Add comment * Serve your private gemma * fix serve yaml * readme * Remove chat completion due to the wrong template * add readme * Update llm/gemma/README.md Co-authored-by: Zongheng Yang <[email protected]> * address comments * Update README.md Co-authored-by: Zongheng Yang <[email protected]> * Update llm/gemma/README.md Co-authored-by: Zongheng Yang <[email protected]> * Update llm/gemma/README.md Co-authored-by: Zongheng Yang <[email protected]> * Update llm/gemma/README.md Co-authored-by: Zongheng Yang <[email protected]> * Change to it * Add chat API * use HF_TOKEN env * typo --------- Co-authored-by: Zongheng Yang <[email protected]>
* when removing cudo credential, sky check fails * remove tips * minor hint fix * fix cluster version for k8s * fix typo
…org#3218) * Add retry for docker pull due to daemon not ready * longer wait time * longer wait time * retry earlier * add retry for retries as well * longer wait time * change wait time * format * Add comment * Fix * Fix indent for azure docker config * Fix docker login config * Fix comments * More robust docker login config * Add retry for docker check * minor fix * Add additional test for stop and start with docker * Fix cancelled
…o hriday/unsloth # Conflicts: # README.md # docs/source/index.rst
This is great, thanks @landscapepainter! I noticed this is only S3/GCS. Do we plan to support Azure as well in this PR? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is very cool @landscapepainter! Left some quick comments.
"""Provides methods to manage and generate Rclone configuration profile.""" | ||
|
||
class RcloneStores(enum.Enum): | ||
S3 = 'S3' |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can we add R2 as well? Should be very similar to S3 except with a different endpoint.
sky/data/mounting_utils.py
Outdated
#f'--log-file {log_file_path} --log-level DEBUG ' #log related flags | ||
# '--dir-cache-time' specifies the frequency of how often rclone should | ||
# check the backend storage for an update when there is a discrepancy. | ||
'--allow-other --vfs-cache-mode writes --dir-cache-time 30s ' |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should we use --vfs-cache-mode full
to also cache reads?
Tested (run the relevant ones):
bash format.sh
sky stop
andsky start
:pytest tests/test_smoke.py::test_aws_storage_mounts_with_stop --aws
pytest tests/test_smoke.py::test_gcp_storage_mounts_with_stop --gcp
sky stop
andsky start
withcomp_test.yaml
to see if all theMOUNT_CACHED
remains storage remains mounted.comp_test.yaml
:launch
tests:sky launch
two instances mounting same backend storage. Create new file on mountpoint on instance 1, and check if it appears shortly in instance 2.spot
jobs tests:launch
on docker containers:comp_test.yaml
:MOUNT_CACHED
mode can recover from preemption to continue training using our vicuna training script.MOUNT_CACHED
mode performs as much as or better than ourMOUNT
mode when writting checkpoints