Implement block copy kernel to optimize beam search #32
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
hongxiayang
pushed a commit
to hongxiayang/vllm
that referenced
this pull request
Feb 13, 2024
slyalin
pushed a commit
to slyalin/vllm
that referenced
this pull request
May 6, 2024
…u8-desc [CPU] Add comment for u8 kvcache layout
tianyil1
pushed a commit
to tianyil1/vllm
that referenced
this pull request
Jun 5, 2024
* add gaudi installation readme * readme writeup * Create README_GAUDI.md * Update README.md * Update README_GAUDI.md * Update README.md * Update readmes
fxmarty
pushed a commit
to fxmarty/vllm-public
that referenced
this pull request
Jun 12, 2024
Update linear.py
joerunde
added a commit
to joerunde/vllm
that referenced
this pull request
Jun 17, 2024
Adds support for multi-lora adapters. Passing tests added over in this PR: https://github.ibm.com/ai-foundation/tgis-deploy-tests/pull/25/files --------- Signed-off-by: Joe Runde <[email protected]>
yukavio
pushed a commit
to yukavio/vllm
that referenced
this pull request
Jul 3, 2024
A warning will be printed out if this case is triggered:
```
WARNING 02-20 22:21:27 sparse_w16a16.py:32] Unstructured sparse kernels are not optimized for NVIDIA SM < 8.0. Naive decompress kernels will be used and can be slower than dense models
```
Works on a T4 with:
```python
from vllm import LLM, SamplingParams
model = LLM(
"nm-testing/opt-125m-pruned2.4",
sparsity="sparse_w16a16",
enforce_eager=True,
dtype="float16",
)
sampling_params = SamplingParams(max_tokens=100, temperature=0)
outputs = model.generate("Hello my name is", sampling_params=sampling_params)
outputs[0].outputs[0].text
```
Test within colab:
https://colab.research.google.com/drive/15xRvWX5gNaTb00BcaXhxwMm6yxavIKGN?usp=sharing
yukavio
pushed a commit
to yukavio/vllm
that referenced
this pull request
Jul 3, 2024
A warning will be printed out if this case is triggered:
```
WARNING 02-20 22:21:27 sparse_w16a16.py:32] Unstructured sparse kernels are not optimized for NVIDIA SM < 8.0. Naive decompress kernels will be used and can be slower than dense models
```
Works on a T4 with:
```python
from vllm import LLM, SamplingParams
model = LLM(
"nm-testing/opt-125m-pruned2.4",
sparsity="sparse_w16a16",
enforce_eager=True,
dtype="float16",
)
sampling_params = SamplingParams(max_tokens=100, temperature=0)
outputs = model.generate("Hello my name is", sampling_params=sampling_params)
outputs[0].outputs[0].text
```
Test within colab:
https://colab.research.google.com/drive/15xRvWX5gNaTb00BcaXhxwMm6yxavIKGN?usp=sharing
bigPYJ1151
pushed a commit
to bigPYJ1151/vllm
that referenced
this pull request
Jul 31, 2024
…ache_weight Enable AWQ to use int8 weight first token as default
Closed
1 task
1 task
1 task
1 task
4 tasks
zyongye
pushed a commit
to zyongye/vllm
that referenced
this pull request
Aug 5, 2025
zyongye
pushed a commit
to zyongye/vllm
that referenced
this pull request
Aug 6, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR implements a block copy kernel. By using this kernel, we can reduce the number of kernel invocations from
2 * num_layers * num_copying_blocksto 1.