Skip to content

[Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell #6479

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 14 commits into from
May 28, 2025
Merged

[Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell #6479

merged 14 commits into from
May 28, 2025

Conversation

Fridge003
Copy link
Collaborator

@Fridge003 Fridge003 commented May 21, 2025 •
edited
Loading

Motivation

Support Blackwell blockwise fp8 kernels supported by flashinfer:
flashinfer-ai/flashinfer#1045
flashinfer-ai/flashinfer#1047

Use environment variable SGLANG_ENABLE_FLASHINFER_GEMM=1 to open flashinfer gemm on blackwell. Also some prior environment variables related to fp8 GEMM are added with SGLANG prefix out of unity.

Flashinfer needs to be pulled from github and installed from source.

Modifications

Checklist

Accuracy Results

SGLANG_ENABLE_FLASHINFER_GEMM=1 python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --attention-backend triton

python3 benchmark/gsm8k/bench_sglang.py --num-shots 8 --num-questions 1319 --parallel 1319
Accuracy: 0.948
Invalid: 0.000
Latency: 48.485 s
Output throughput: 2947.232 token/s

Profile Results

SGLANG_ENABLE_FLASHINFER_GEMM=1 python3 -m sglang.bench_one_batch --model-path deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --attention-backend triton --batch 16 --input-len 1024 --output-len 2 --profile

Before:
截屏2025-05-28 00 47 56

After:
截屏2025-05-28 00 48 56

Benchmark Results

python3 -m sglang.bench_serving --backend sglang-oai  --dataset-name random --random-input-len 1000 --random-output-len 1000 --random-range-ratio 1 --num-prompts 128 --max-concurrency 16

Before:

============ Serving Benchmark Result ============
Backend:                                 sglang-oai
Traffic request rate:                    inf       
Max request concurrency:                 16        
Successful requests:                     128       
Benchmark duration (s):                  191.30    
Total input tokens:                      128000    
Total generated tokens:                  128000    
Total generated tokens (retokenized):    127507    
Request throughput (req/s):              0.67      
Input token throughput (tok/s):          669.12    
Output token throughput (tok/s):         669.12    
Total token throughput (tok/s):          1338.24   
Concurrency:                             15.99     
----------------End-to-End Latency----------------
Mean E2E Latency (ms):                   23901.69  
Median E2E Latency (ms):                 22786.18  
---------------Time to First Token----------------
Mean TTFT (ms):                          1790.58   
Median TTFT (ms):                        978.69    
P99 TTFT (ms):                           8854.51   
---------------Inter-Token Latency----------------
Mean ITL (ms):                           22.24     
Median ITL (ms):                         21.79     
P95 ITL (ms):                            22.75     
P99 ITL (ms):                            23.47     
Max ITL (ms):                            4906.84   
==================================================

After:

============ Serving Benchmark Result ============
Backend:                                 sglang-oai
Traffic request rate:                    inf       
Max request concurrency:                 16        
Successful requests:                     128       
Benchmark duration (s):                  186.06    
Total input tokens:                      128000    
Total generated tokens:                  128000    
Total generated tokens (retokenized):    127307    
Request throughput (req/s):              0.69      
Input token throughput (tok/s):          687.97    
Output token throughput (tok/s):         687.97    
Total token throughput (tok/s):          1375.94   
Concurrency:                             15.99     
----------------End-to-End Latency----------------
Mean E2E Latency (ms):                   23246.34  
Median E2E Latency (ms):                 22506.67  
---------------Time to First Token----------------
Mean TTFT (ms):                          1408.12   
Median TTFT (ms):                        924.91    
P99 TTFT (ms):                           6815.20   
---------------Inter-Token Latency----------------
Mean ITL (ms):                           21.99     
Median ITL (ms):                         21.58     
P95 ITL (ms):                            22.58     
P99 ITL (ms):                            23.35     
Max ITL (ms):                            5320.55   
==================================================

@Fridge003
Copy link
Collaborator Author

cc @zhyncs @yzh119

@Fridge003 Fridge003 changed the title [Feature] Support Flashinfer fp8 blocakwise GEMM kernel on Blackwell [Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell May 21, 2025
@Fridge003 Fridge003 requested a review from zhaochenyang20 as a code owner May 21, 2025 06:50
@zhyncs zhyncs self-assigned this May 21, 2025
yongwww
yongwww approved these changes May 22, 2025
View reviewed changes
Copy link

@yongwww yongwww left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks good.

@Fridge003 Fridge003 mentioned this pull request May 22, 2025
Closed
2 tasks
b8zhong
b8zhong reviewed May 25, 2025
View reviewed changes
Copy link
Collaborator

@b8zhong b8zhong left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should the env var be prefixed with SGLANG_?

Just a suggestion

@Fridge003
Copy link
Collaborator Author

Should the env var be prefixed with SGLANG_?

Just a suggestion

Nice suggestion! Just modified.

zhyncs
zhyncs requested changes May 28, 2025
View reviewed changes
zhyncs
zhyncs reviewed May 28, 2025
View reviewed changes
@zhyncs zhyncs merged commit 791b3bf into sgl-project:main May 28, 2025
1 of 36 checks passed
@Fridge003 Fridge003 mentioned this pull request May 28, 2025
Merged
6 tasks
@Fridge003 Fridge003 deleted the kernel branch May 28, 2025 23:09
ChangyiYang pushed a commit to ChangyiYang/sglang-changyi that referenced this pull request May 29, 2025
@Fridge003 @ChangyiYang
[Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell (s…
7df4699 
…gl-project#6479)
@Fridge003
Copy link
Collaborator Author

Waiting for flashinfer-ai/flashinfer#1102 to merge,
then the contiguous operations can be removed

@Fridge003 Fridge003 mentioned this pull request Jun 2, 2025
Merged
6 tasks
Layssy pushed a commit to Layssy/sglang-iaas that referenced this pull request Jun 9, 2025
@Fridge003 @Layssy
[Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell (s…
6652432 
…gl-project#6479)
xwu-intel pushed a commit to xwu-intel/sglang that referenced this pull request Jun 17, 2025
@Fridge003 @xwu-intel
[Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell (s…
1ebd069 
…gl-project#6479)
walker-ai pushed a commit to walker-ai/sglang that referenced this pull request Jul 8, 2025
@yudian0504
PullRequest: 52 sgl_20250610_sync_tag047
a19e2e2 
Merge branch 'sgl_20250610_sync_tag047 of git@code.alipay.com:Theta/SGLang.git into main

https://code.alipay.com/Theta/SGLang/pull_requests/52


Reviewed-by: 剑川 <jianchuan.gys@antgroup.com>


* [Bugfix] Fix slice operation when chunk size mismatch (sgl-project#6697)
* [Bugfix] Fix ChatCompletion endpoint of mini_lb when stream is set (sgl-project#6703)
* [CI] Fix setup of disaggregation with different tp (sgl-project#6706)
* [PD] Remove Unnecessary Exception Handling for FastQueue.get() (sgl-project#6712)
* Fuse routed_scaling_factor in DeepSeek (sgl-project#6710)
* Overlap two kernels in DeepSeek with communication (sgl-project#6711)
* Minor refactor two-batch overlap (sgl-project#6682)
* Speed up when having padding tokens two-batch overlap (sgl-project#6668)
* [Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell (sgl-project#6479)
* Fix LoRA bench (sgl-project#6719)
* temp
* Fix PP for Qwen3 MoE (sgl-project#6709)
* [feat] triton kernel for get_last_loc (sgl-project#6676)
* [fix] more mem for draft_extend cuda_graph (sgl-project#6726)
* [PD] bug fix:  Update status if nixl receiver send a a dummy req. (sgl-project#6720)
* Tune memory arguments on B200 (sgl-project#6718)
* Add DeepSeek-R1-0528 function call chat template (sgl-project#6725)
* refactor(tool call): Fix BaseFormatDetector tool_index issue and refactor `parse_streaming_increment` (sgl-project#6715)
* Add draft extend CUDA graph for Triton backend (sgl-project#6705)
* refactor apply_w8a8_block_fp8_linear in fp (sgl-project#6545)
* [PD] Support completion endpoint (sgl-project#6729)
* PD Rust LB (PO2) (sgl-project#6437)
* Super tiny enable sole usage of expert distribution metrics and update doc (sgl-project#6680)
* Support picking variants of EPLB algorithms (sgl-project#6728)
* Support tuning DeepEP configs (sgl-project#6742)
* [test] add ut and bm for get_last_loc (sgl-project#6746)
* Fix mem_fraction_static for AMD CI (sgl-project#6748)
* [fix][RL] Fix DeepSeekV3ForCausalLM.post_load_weights for multiple update weight (sgl-project#6265)
* Improve EPLB logical to physical dispatch map (sgl-project#6727)
* Update DeepSeek-R1-0528 function call chat template (sgl-project#6765)
* [PD] Optimize time out logic and add env var doc for mooncake (sgl-project#6761)
* Fix aiohttp 'Chunk too big' in bench_serving (sgl-project#6737)
* Support sliding window in triton backend (sgl-project#6509)
* Fix shared experts fusion error (sgl-project#6289)
* Fix one bug in the grouped-gemm triton kernel (sgl-project#6772)
* update llama4 chat template and pythonic parser (sgl-project#6679)
* feat(tool call): Enhance Llama32Detector for improved JSON parsing in non-stream (sgl-project#6784)
* Support token-level quantization for EP MoE (sgl-project#6782)
* Temporarily lower mmlu threshold for triton sliding window backend (sgl-project#6785)
* ci: relax test_function_call_required (sgl-project#6786)
* Add intel_amx backend for Radix Attention for CPU (sgl-project#6408)
* Fix incorrect LoRA weight loading for fused gate_up_proj (sgl-project#6734)
* fix(PD-disaggregation): Can not get local ip (sgl-project#6792)
* [FIX] mmmu bench serving result display error (sgl-project#6525) (sgl-project#6791)
* Bump torch to 2.7.0 (sgl-project#6788)
* chore: bump sgl-kernel v0.1.5 (sgl-project#6794)
* Improve profiler and integrate profiler in bench_one_batch_server (sgl-project#6787)
* chore: upgrade sgl-kernel v0.1.5 (sgl-project#6795)
* [Minor] Always append newline after image token when parsing chat message (sgl-project#6797)
* Update CI tests for Llama4 models (sgl-project#6421)
* [Feat] Enable PDL automatically on Hopper architecture (sgl-project#5981)
* chore: update blackwell docker (sgl-project#6800)
* misc: cache is_hopper_arch (sgl-project#6799)
* Remove contiguous before Flashinfer groupwise fp8 gemm (sgl-project#6804)
* Correctly abort the failed grammar requests & Improve the handling of abort (sgl-project#6803)
* [EP] Add cuda kernel for moe_ep_pre_reorder (sgl-project#6699)
* Add draft extend CUDA graph for flashinfer backend  (sgl-project#6805)
* Refactor CustomOp to avoid confusing bugs (sgl-project#5382)
* Tiny log prefill time (sgl-project#6780)
* Tiny fix EPLB assertion about rebalancing period and recorder window size (sgl-project#6813)
* Add simple utility to dump tensors for debugging (sgl-project#6815)
* Fix profiles do not have consistent names (sgl-project#6811)
* Speed up rebalancing when using non-static dispatch algorithms (sgl-project#6812)
* [1/2] Add Kernel support for Cutlass based Fused FP4 MoE (sgl-project#6093)
* [Router] Fix k8s Service Discovery (sgl-project#6766)
* Add CPU optimized kernels for topk and rope fusions  (sgl-project#6456)
* fix new_page_count_next_decode (sgl-project#6671)
* Fix wrong weight reference in dynamic EPLB (sgl-project#6818)
* Minor add metrics to expert location updater (sgl-project#6816)
* [Refactor] Rename `n_share_experts_fusion` as `num_fused_shared_experts` (sgl-project#6735)
* [FEAT] Add transformers backend support  (sgl-project#5929)
* [fix] recover auto-dispatch for rmsnorm and rope (sgl-project#6745)
* fix ep_moe_reorder kernel bugs (sgl-project#6858)
* [Refactor] Multimodal data processing for VLM (sgl-project#6659)
* Decoder-only Scoring API (sgl-project#6460)
* feat: add dp-rank to KV events (sgl-project#6852)
* Set `num_fused_shared_experts` as `num_shared_experts` when shared_experts fusion is not disabled (sgl-project#6736)
* Fix one missing arg in DeepEP (sgl-project#6878)
* Support LoRA in TestOpenAIVisionServer and fix fused kv_proj loading bug. (sgl-project#6861)
* support 1 shot allreduce  in 1-node and 2-node using mscclpp (sgl-project#6277)
* Fix Qwen3MoE missing token padding optimization (sgl-project#6820)
* Tiny update error hints (sgl-project#6846)
* Support layerwise rebalancing experts (sgl-project#6851)
* Tiny allow profiler API to auto create directory (sgl-project#6865)
* Support Blackwell DeepEP docker images (sgl-project#6868)
* [EP] Add cuda kernel for moe_ep_post_reorder (sgl-project#6837)
* [theta]merge 0605
* oai: fix openAI client error with single request via batch api (sgl-project#6170)
* [PD] Fix potential perf spike caused by tracker gc and optimize doc (sgl-project#6764)
* Use deepgemm instead of triton for fused_qkv_a_proj_with_mqa (sgl-project#6890)
* [CUTLASS-FP4-MOE]  Introduce CutlassMoEParams class for easy initialization of Cutlass Grouped Gems Metadata (sgl-project#6887)
* bugfix(OAI): Fix image_data processing for jinja chat templates (sgl-project#6877)
* [CPU] enable CI for PRs, add Dockerfile and auto build task (sgl-project#6458)
* AITER backend extension and workload optimizations (sgl-project#6838)
* [theta]merge
* [theta]merge
* [Feature] Support Flashinfer fmha on Blackwell (sgl-project#6930)
* Fix a bug in abort & Improve docstrings for abort (sgl-project#6931)
* Tiny support customize DeepEP max dispatch tokens per rank (sgl-project#6934)
* Sync the changes on cuda graph runners (sgl-project#6932)
* [PD] Optimize transfer queue forward logic for dummy rank (sgl-project#6922)
* [Refactor] image data process in bench_serving (sgl-project#6879)
* [fix] logical_to_all_physical_map index 256 is out of bounds in EP parallel. (sgl-project#6767)
* Add triton fused moe kernel config for E=257 on B200 (sgl-project#6939)
* [sgl-kernel] update deepgemm (sgl-project#6942)
* chore: bump sgl-kernel v0.1.6 (sgl-project#6943)
* Minor compile fused topk (sgl-project#6944)
* [Bugfix] pipeline parallelism and Eagle Qwen2 (sgl-project#6910)
* Tiny re-introduce profile id logging (sgl-project#6912)
* Add triton version as a fused_moe_triton config search key to avoid performace decrease in different Triton version (sgl-project#5955)
* reduce torch.zeros overhead in moe align block size kernel (sgl-project#6369)
* chore: upgrade sgl-kernel v0.1.6 (sgl-project#6945)
* add fbgemm moe grouped gemm kernel benchmark (sgl-project#6924)
* [Docker] Add docker file for SGL Router (sgl-project#6915)
* Disabling mixed chunked prefill when eagle is enabled (sgl-project#6874)
* Add canary for EPLB rebalancing (sgl-project#6895)
* Refactor global_server_args_dict (sgl-project#6866)
* Fuse routed scaling factor in topk_reduce kernel (sgl-project#6220)
* Update server timeout time in AMD CI. (sgl-project#6953)
* [misc] add is_cpu() (sgl-project#6950)
* Add H20 fused MoE kernel tuning configs for DeepSeek-R1/V3 (sgl-project#6885)
* Add a CUDA kernel for fusing mapping and weighted sum for MoE. (sgl-project#6916)
* chore: bump sgl-kernel v0.1.6.post1 (sgl-project#6955)
* chore: upgrade sgl-kernel v0.1.6.post1 (sgl-project#6957)
* [DeepseekR1-FP4] Add Support for nvidia/DeepSeekR1-FP4 model (sgl-project#6853)
* Revert "Fuse routed scaling factor in topk_reduce kernel (sgl-project#6220)" (sgl-project#6968)
* [AMD] Add more tests to per-commit-amd (sgl-project#6926)
* chore: bump sgl-kernel v0.1.7 (sgl-project#6963)
* Slightly improve the sampler to skip unnecessary steps (sgl-project#6956)
* rebase h20 fused_moe config (sgl-project#6966)
* Fix CI and triton moe Configs (sgl-project#6974)
* Remove unnecessary kernels of num_token_non_padded (sgl-project#6965)
* Extend cuda graph capture bs for B200 (sgl-project#6937)
* Fuse routed scaling factor in deepseek (sgl-project#6970)
* Sync cuda graph runners (sgl-project#6976)
* Fix draft extend ut stability with flush cache (sgl-project#6979)
* Fix triton sliding window test case (sgl-project#6981)
* Fix expert distribution dumping causes OOM (sgl-project#6967)
* Minor remove one kernel for DeepSeek (sgl-project#6977)
* [perf][sgl-kernel] extend cutlass_mla_decode to support num_head < 128 (sgl-project#6929)
* Enable more unit tests for AMD CI. (sgl-project#6983)
* Use torch.compile to fuse flash attention decode metadata preparation (sgl-project#6973)
* Eliminate stream sync to speed up LoRA batch init  (sgl-project#6960)
* support qwen3 emebedding (sgl-project#6990)
* Fix torch profiler bugs for bench_offline_throughput.py (sgl-project#6557)
* chore: upgrade flashinfer v0.2.6.post1 jit (sgl-project#6958)
* cleanup tmp dir (sgl-project#7007)
* chore: update pr test xeon (sgl-project#7008)
* Fix cutlass MLA gets almost zero accuracy (sgl-project#6998)
* Update amd nightly models CI. (sgl-project#6992)
* feat: add direct routing strategy to DP worker (sgl-project#6884)
* Fallback to lower triton version for unfound fused moe configs (sgl-project#7013)
* Fix torchvision version for Blackwell (sgl-project#7015)
* Simplify prepare_extend_after_decode (sgl-project#6987)
* Migrate to assertEqual (sgl-project#6741)
* Fix torch version in blackwell dockerfile (sgl-project#7017)
* chore: update pr test xeon (sgl-project#7018)
* Update default settings for blackwell (sgl-project#7023)
* Support both approximate and exact expert distribution collection (sgl-project#6964)
* Add decode req pool (sgl-project#6980)
* [theta]merge 0610
* [theta]merge 0610
* [CI] Add CI workflow for sgl-router docker build (sgl-project#7027)
* Fix fused_moe triton configs (sgl-project#7029)
* CPU: map changes from developing branch in sgl-kernel (sgl-project#6833)
* chore: bump v0.4.7 (sgl-project#7038)
* Update README.md (sgl-project#7040)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@zhyncs zhyncs zhyncs approved these changes

@merrymercy merrymercy Awaiting requested review from merrymercy merrymercy is a code owner

@Ying1123 Ying1123 Awaiting requested review from Ying1123 Ying1123 is a code owner

@ispobock ispobock Awaiting requested review from ispobock ispobock is a code owner

@HaiShaw HaiShaw Awaiting requested review from HaiShaw HaiShaw is a code owner

@ch-wan ch-wan Awaiting requested review from ch-wan ch-wan is a code owner

@BBuf BBuf Awaiting requested review from BBuf BBuf is a code owner

@zhaochenyang20 zhaochenyang20 Awaiting requested review from zhaochenyang20

+2 more reviewers

@b8zhong b8zhong b8zhong left review comments

@yongwww yongwww yongwww approved these changes

Reviewers whose approvals may not affect merge requirements
Assignees

@zhyncs zhyncs

Labels
high priority
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@Fridge003 @yongwww @zhyncs @b8zhong