- 18 Jul, 2024 1 commit
-
-
zhuwenwen authored
-
- 17 Jul, 2024 1 commit
-
-
zhuwenwen authored
-
- 12 Jul, 2024 1 commit
-
-
zhuwenwen authored
-
- 09 Jul, 2024 1 commit
-
-
huangwb authored
-
- 08 Jul, 2024 1 commit
-
-
zhuwenwen authored
-
- 06 Jul, 2024 2 commits
- 01 Jul, 2024 1 commit
-
-
zhuwenwen authored
-
- 28 Jun, 2024 1 commit
-
-
zhuwenwen authored
-
- 20 Jun, 2024 1 commit
-
-
zhuwenwen authored
-
- 11 Jun, 2024 9 commits
-
-
Nick Hill authored
Co-authored-by:Antoni Baum <antoni.baum@protonmail.com>
-
youkaichao authored
-
sasha0552 authored
-
Ali Panahi authored
-
Junichi Sato authored
-
Woosuk Kwon authored
-
Cyrus Leung authored
-
maor-ps authored
Co-authored-by:DarkLight1337 <tlleungac@connect.ust.hk>
-
Nick Hill authored
-
- 10 Jun, 2024 8 commits
-
-
Simon Mo authored
-
Cyrus Leung authored
-
Itay Etelis authored
-
Cyrus Leung authored
Co-authored-by:Roger Wang <ywang@roblox.com>
-
Cyrus Leung authored
-
youkaichao authored
-
Roger Wang authored
-
Dipika Sikka authored
Co-authored-by:Michael Goin <michael@neuralmagic.com>
-
- 09 Jun, 2024 3 commits
-
-
Bla_ckB authored
-
bnellnm authored
-
youkaichao authored
[Core][CUDA Graph] add output buffer for cudagraph to reduce memory footprint (#5074)
-
- 08 Jun, 2024 3 commits
-
-
Michael Goin authored
-
Hongxia Yang authored
-
Cheng Li authored
Bug description: With torch 2.4.0.dev20240603+cu121, cutlass_fp8_supported outputs False, and the (capability, version) before the comparison is (90, 11111111112) This PR fixes the support check for FP8 CUTLASS ( cutlass_fp8_supported) which was introduced in https://github.com/vllm-project/vllm/pull/5183.
-
- 07 Jun, 2024 7 commits
-
-
Calvinn Ng authored
Co-authored-by:team <calvinn.ng@ahrefs.com>
-
Roger Wang authored
Co-authored-by: -
Dipika Sikka authored
Co-authored-by:
Varun Sundar Rabindranath <varunsundar08@gmail.com> Co-authored-by:
Varun Sundar Rabindranath <varun@neuralmagic.com>
-
limingshu authored
-
Antoni Baum authored
-
Tyler Michael Smith authored
Switching from torch._scaled_mm to vLLM's cutlass fp8 kernels when supported as we are seeing 5-15% improvement in e2e performance on neuralmagic/Meta-Llama-3-8B-Instruct-FP8 see https://docs.google.com/spreadsheets/d/1GiAnmzyGHgZ6zL_LDSTm35Bdrt4A8AaFEurDlISYYA4/ for some quick e2e benchmarks and #5144 for comparisons across different GEMM sizes.
-
youkaichao authored
-
