[NPU] Lm_head slice -> split #12231
Open
[NPU] Lm_head slice -> split #12231
Conversation
This file contains 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
| @@ -94,9 +94,9 @@ def optimize_llm_pre(model: torch.nn.Module, qtype, mixed_precision): | |||
| not cpu_lm_head: | |||
| # Do not split lm_head and use sym_int8 instead when mixed_precison is True | |||
| is_split = (not mixed_precision) and qtype == "sym_int4_rtn" | |||
| split_num = 14 if is_split else 1 | |||
| split_num = 28 if is_split else 1 | |||
There was a problem hiding this comment.
Should we make split_num=14 if use_split=False?
There was a problem hiding this comment.
I'm considering whether we should expose the use_split option to users.
Y (make split_num=14 if use_split=False)
- Pros: We can recommend the users set
use_split=Falseto avoid the performance degradation on driver versions below 2715 - Cons: It may need to be clarified for the user.
N (should update the readme recommended driver section)
- Pros: This approach is more straightforward for users, and less configuration is needed.
- Cons: Users must ensure they use driver version 2908 or higher. For MTL users, we would need to fall back to split_num=14 w/ slice.
Please let me know if you have any comments. @Oscilloscope98 @jason-dai
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Description
lm_head from slice linears -> split dq linears.
1. Why the change?
2. User API changes
3. Summary of the change
4. How to test?
1234). And paste your action link here once it has been successfully finished.5. New dependencies
- Dependency1
- Dependency2
- ...
- Dependency1 and license1
- Dependency2 and license2
- ...