ln + fp8 quant benchmark #109765
Closed
ln + fp8 quant benchmark #109765
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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/109765
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit b78a428 with merge base 59592ce ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Benchmark results: ``` Config: float8_dtype=torch.float8_e4m3fn, shape=(4, 2048, 4096), enable_split_reductions=False. Benchmark results: Inductor: 12.767714285714286ms, Eager: 0.8525048543689321ms. Config: float8_dtype=torch.float8_e4m3fn, shape=(4, 2048, 4096), enable_split_reductions=True. Benchmark results: Inductor: 0.11353932584269663ms, Eager: 0.8525339805825243ms. Config: float8_dtype=torch.float8_e5m2, shape=(4, 2048, 4096), enable_split_reductions=False. Benchmark results: Inductor: 12.859285714285713ms, Eager: 0.8535865384615384ms. Config: float8_dtype=torch.float8_e5m2, shape=(4, 2048, 4096), enable_split_reductions=True. Benchmark results: Inductor: 0.11312808988764046ms, Eager: 0.8531844660194176ms. ``` When "split_reductions" is disabled, I expect that there is only one fused kernel. However now there are still two kernels (for LN and amax calculation). This is the generated code: https://gist.github.com/ipiszy/6ab2c86ba211240d606edfab8b14e7bd. Kernel 1: block size: {XBLOCK: 1, RBLOCK: 4096}. Kernel 2: block size: {XBLOCK: 1, RBLOCK: 2048} I think ideally, there should be 2 kernels: the first kernel does block-level amax calculation, and the second kernel aggregates amax results from the first kernel. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ngimel yf225 chenyang78 kadeng muchulee8 aakhundov [ghstack-poisoned]
Benchmark results: ``` Config: float8_dtype=torch.float8_e4m3fn, shape=(4, 2048, 4096), enable_split_reductions=False. Benchmark results: Inductor: 12.767714285714286ms, Eager: 0.8525048543689321ms. Config: float8_dtype=torch.float8_e4m3fn, shape=(4, 2048, 4096), enable_split_reductions=True. Benchmark results: Inductor: 0.11353932584269663ms, Eager: 0.8525339805825243ms. Config: float8_dtype=torch.float8_e5m2, shape=(4, 2048, 4096), enable_split_reductions=False. Benchmark results: Inductor: 12.859285714285713ms, Eager: 0.8535865384615384ms. Config: float8_dtype=torch.float8_e5m2, shape=(4, 2048, 4096), enable_split_reductions=True. Benchmark results: Inductor: 0.11312808988764046ms, Eager: 0.8531844660194176ms. ``` When "split_reductions" is disabled, I expect that there is only one fused kernel. However now there are still two kernels (for LN and amax calculation). This is the generated code: https://gist.github.com/ipiszy/6ab2c86ba211240d606edfab8b14e7bd. Kernel 1: block size: {XBLOCK: 1, RBLOCK: 4096}. Kernel 2: block size: {XBLOCK: 1, RBLOCK: 2048} I think ideally, there should be 2 kernels: the first kernel does block-level amax calculation, and the second kernel aggregates amax results from the first kernel. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ngimel yf225 chenyang78 kadeng muchulee8 aakhundov [ghstack-poisoned]
Closed
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.
Benchmark results:
When "split_reductions" is disabled, I expect that there is only one fused kernel. However now there are still two kernels (for LN and amax calculation). This is the generated code: https://gist.github.com/ipiszy/6ab2c86ba211240d606edfab8b14e7bd. Kernel 1: block size: {XBLOCK: 1, RBLOCK: 4096}. Kernel 2: block size: {XBLOCK: 1, RBLOCK: 2048}
I think ideally, there should be 2 kernels: the first kernel does block-level amax calculation, and the second kernel aggregates amax results from the first kernel.
Stack from ghstack (oldest at bottom):
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @yf225 @chenyang78 @kadeng @muchulee8 @aakhundov @ColinPeppler @ngimel