Update vLLM to 0.6.6.post1#2892
Update vLLM to 0.6.6.post1#2892mergify[bot] merged 1 commit intoinstructlab:maininstructlab/instructlab:mainfrom fabiendupont:update-vllm-0.6.6.post1fabiendupont/instructlab:update-vllm-0.6.6.post1Copy head branch name to clipboard
Conversation
8fe682b to
d09d8f3
Compare
|
We'll need a new SDG z-stream that allows for a higher Torch version to resolve the dependency issue that will allow us to test this - that backport is being worked on here: instructlab/sdg#469 |
|
SDG v0.6.3 has been released - the only change from v0.6.2 (used by instructlab currently) is the raising of this PyTorch upper version limit, which should unblock this PR. |
|
@fabiendupont can you rebase this? we'll see if CI is able to complete the install and then we can test if the new vLLM package will work |
d09d8f3 to
1d9654e
Compare
|
This pull request has merge conflicts that must be resolved before it can be |
1d9654e to
7dcdccf
Compare
|
Do we need to manually kick off a large or xlarge CI test for this so that vllm actually gets used? I think both the small and medium e2e tests use llama_cpp, right? |
Small is using only llama_cpp, Medium actually uses both: https://github.com/instructlab/instructlab/blob/main/docs/ci.md/#e2e-test-coverage-matrix Now |
|
E2E (NVIDIA L40S x4) workflow launched on this PR: View run |
|
e2e workflow succeeded on this PR: View run, congrats! |
|
We have some users using a combo with 0.6.4.post1, would love to see a CI test run pass on that. |
Do you think we should be updating to that version instead? If not, can we run that in a separate PR so we can get this in as early before the 0.23.0 release as we can? |
As an intermediate step to updating vLLM to 0.6.6.post1 in instructlab#2892, this pull request updates vLLM to 0.6.4.post1 to ensure that the CI tests succeed with this version too. Signed-off-by: Fabien Dupont <fdupont@redhat.com>
|
@n1hility, @nathan-weinberg, I have created #2910 that updates vLLM to 0.6.4.post1. This will allow us to run the CI tests and be more comfortable with the bigger leap forward to 0.6.6.post1. |
As an intermediate step to updating vLLM to 0.6.6.post1 in #2892, this pull request updates vLLM to 0.6.4.post1 to ensure that the CI tests succeed with this version too. Approved-by: nathan-weinberg Approved-by: n1hility
As an intermediate step to updating vLLM to 0.6.6.post1 in instructlab#2892, this pull request updates vLLM to 0.6.4.post1 to ensure that the CI tests succeed with this version too. Signed-off-by: Fabien Dupont <fdupont@redhat.com>
|
This pull request has merge conflicts that must be resolved before it can be |
As an intermediate step to updating vLLM to 0.6.6.post1 in instructlab#2892, this pull request updates vLLM to 0.6.4.post1 to ensure that the CI tests succeed with this version too. Approved-by: nathan-weinberg Approved-by: n1hility
7dcdccf to
0e60b4d
Compare
0e60b4d to
e3574f2
Compare
This PR is a follow-up to instructlab#2865 that relaxed the PyTorch version range. Even with that range extension, we realized that PyTorch 2.4 is still used when installing `instructlab[vllm-cuda]`, because vLLM 0.6.2 has a requirement on PyTorch 2.4. This new PR updates the version of vLLM to 0.6.6.post1, which is the latest available in the Open Data Hub fork of vLLM. The vLLM changelog doesn't highlight much risk in this version bump. Resolves instructlab#2702 Signed-off-by: Fabien Dupont <fdupont@redhat.com>
e3574f2 to
9c4a31c
Compare
|
E2E (NVIDIA L40S x4) workflow launched on this PR: View run |
|
e2e workflow succeeded on this PR: View run, congrats! |
As an intermediate step to updating vLLM to 0.6.6.post1 in instructlab#2892, this pull request updates vLLM to 0.6.4.post1 to ensure that the CI tests succeed with this version too. Signed-off-by: Fabien Dupont <fdupont@redhat.com> Signed-off-by: Kevin Dubois <kevin.dubois@gmail.com>
This PR is a follow-up to #2865 that relaxed the PyTorch version range. Even with that range extension, we realized that PyTorch 2.4 is still used when installing
instructlab[vllm-cuda], because vLLM 0.6.2 has a requirement on PyTorch 2.4.This new PR updates the version of vLLM to 0.6.6.post1, which is the latest available in the Open Data Hub fork of vLLM. The vLLM changelog doesn't highlight much risk in this version bump.
Resolves #2702