[JIT sgl-kernel] Jit support per tensor quant#15709
Merged
Conversation
This reverts commit 4b1b88d.
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
5f32ae5 to
fea8cb7
Compare
FlamingoPg
reviewed
Dec 25, 2025
FlamingoPg
approved these changes
Dec 25, 2025
Leoyzen
pushed a commit
to Leoyzen/sglang
that referenced
this pull request
Dec 25, 2025
Leoyzen
pushed a commit
to Leoyzen/sglang
that referenced
this pull request
Dec 25, 2025
YChange01
pushed a commit
to YChange01/sglang
that referenced
this pull request
Jan 13, 2026
21 tasks
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Motivation
jit_kernel python3 /home/lmsys/bbuf/sglang/sgl-kernel/benchmark/bench_per_tensor_quant_fp8.py ⚠️ vLLM not available, skipping comparison per-tensor-quant-fp8-performance: batch_size seq_len SGL Kernel 0 16.0 64.0 10.249762 1 16.0 128.0 21.541752 2 16.0 256.0 48.441215 3 16.0 512.0 106.394665 4 16.0 1024.0 206.246530 5 16.0 2048.0 403.524995 6 32.0 64.0 21.571416 7 32.0 128.0 48.642592 8 32.0 256.0 106.193720 9 32.0 512.0 206.433368 10 32.0 1024.0 403.411667 11 32.0 2048.0 884.436687 12 64.0 64.0 48.301857 13 64.0 128.0 106.155326 14 64.0 256.0 206.425190 15 64.0 512.0 403.736333 16 64.0 1024.0 884.741346 17 64.0 2048.0 1761.169354 18 128.0 64.0 105.996946 19 128.0 128.0 206.263826 20 128.0 256.0 403.728982 21 128.0 512.0 884.528001 22 128.0 1024.0 1761.481285 23 128.0 2048.0 3514.272054vLLM not available, skipping comparison per-tensor-quant-fp8-performance: batch_size seq_len SGL Kernel 0 16.0 64.0 10.227294 1 16.0 128.0 21.485734 2 16.0 256.0 48.617873 3 16.0 512.0 106.206471 4 16.0 1024.0 206.348592 5 16.0 2048.0 403.680662 6 32.0 64.0 21.730780 7 32.0 128.0 49.148478 8 32.0 256.0 106.192177 9 32.0 512.0 206.320342 10 32.0 1024.0 403.442999 11 32.0 2048.0 884.408673 12 64.0 64.0 48.800135 13 64.0 128.0 106.235504 14 64.0 256.0 206.406806 15 64.0 512.0 403.754671 16 64.0 1024.0 884.618640 17 64.0 2048.0 1760.879993 18 128.0 64.0 106.232718 19 128.0 128.0 206.548824 20 128.0 256.0 403.857668 21 128.0 512.0 884.276668 22 128.0 1024.0 1760.829290 23 128.0 2048.0 3514.077346Modifications
Accuracy Tests
Benchmarking and Profiling
Checklist