- 30 Jul, 2024 3 commits
-
-
Varun Sundar Rabindranath authored
Co-authored-by:Varun Sundar Rabindranath <varun@neuralmagic.com>
-
Tyler Michael Smith authored
-
Tyler Michael Smith authored
-
- 29 Jul, 2024 2 commits
-
-
Tyler Michael Smith authored
-
Varun Sundar Rabindranath authored
Co-authored-by:
-
- 27 Jul, 2024 2 commits
-
-
Alexander Matveev authored
-
Lucas Wilkinson authored
-
- 26 Jul, 2024 1 commit
-
-
Tyler Michael Smith authored
-
- 22 Jul, 2024 1 commit
-
-
Tyler Michael Smith authored
-
- 21 Jul, 2024 1 commit
-
-
Alexander Matveev authored
-
- 20 Jul, 2024 1 commit
-
-
Varun Sundar Rabindranath authored
Co-authored-by:
-
- 18 Jul, 2024 1 commit
-
-
Varun Sundar Rabindranath authored
Co-authored-by:
-
- 14 Jul, 2024 1 commit
-
-
Tyler Michael Smith authored
-
- 03 Jul, 2024 1 commit
-
-
Michael Goin authored
-
- 28 Jun, 2024 1 commit
-
-
Tyler Michael Smith authored
-
- 26 Jun, 2024 1 commit
-
-
Luka Govedič authored
Co-authored-by:
Chih-Chieh-Yang <7364402+cyang49@users.noreply.github.com> Co-authored-by:
Lucas Wilkinson <lwilkinson@neuralmagic.com>
-
- 23 Jun, 2024 1 commit
-
-
Varun Sundar Rabindranath authored
Co-authored-by:
-
- 20 Jun, 2024 3 commits
-
-
Tyler Michael Smith authored
-
Varun Sundar Rabindranath authored
Co-authored-by: -
Varun Sundar Rabindranath authored
Co-authored-by:
-
- 18 Jun, 2024 1 commit
-
-
Tyler Michael Smith authored
-
- 14 Jun, 2024 2 commits
-
-
Tyler Michael Smith authored
-
Tyler Michael Smith authored
-
- 13 Jun, 2024 1 commit
-
-
Tyler Michael Smith authored
Co-authored-by:
Michael Goin <michael@neuralmagic.com> Co-authored-by:
youkaichao <youkaichao@gmail.com> Co-authored-by:
zifeitong <zifei.tong@parasail.io> Co-authored-by:
Robert Shaw <114415538+robertgshaw2-neuralmagic@users.noreply.github.com>
-
- 12 Jun, 2024 1 commit
-
-
Cody Yu authored
Inspired by #5146, this PR improves FP8 quantize kernel by vectorizing data transfer to better utilize memory bandwidth. Microbenchmark shows that this improved kernel can achieve 1.0x-1.5x speedup (especially when hidden size is large). In details, we applied 3 optimizations: - Use inverted scale so that most divisions are changed to multiplications. - Unroll the loop by 4 times to improve ILP. - Use vectorized 4 to transfer data between HBM and SRAM.
-
- 09 Jun, 2024 1 commit
-
-
bnellnm authored
-
- 07 Jun, 2024 1 commit
-
-
Dipika Sikka authored
Co-authored-by:
Varun Sundar Rabindranath <varunsundar08@gmail.com> Co-authored-by:
-
- 05 Jun, 2024 1 commit
-
-
Tyler Michael Smith authored
Co-authored-by:Cody Yu <hao.yu.cody@gmail.com>
-
- 03 Jun, 2024 1 commit
-
-
Tyler Michael Smith authored
-
- 01 Jun, 2024 3 commits
-
-
Varun Sundar Rabindranath authored
Co-authored-by:
-
Tyler Michael Smith authored
-
Tyler Michael Smith authored
-
- 31 May, 2024 2 commits
-
-
Simon Mo authored
Revert "[Kernel] Marlin_24: Ensure the mma.sp instruction is using the ::ordered_metadata modifier (introduced with PTX 8.5)" (#5149)
-
Alexander Matveev authored
[Kernel] Marlin_24: Ensure the mma.sp instruction is using the ::ordered_metadata modifier (introduced with PTX 8.5) (#5136)
-
- 23 May, 2024 2 commits
-
-
Dipika Sikka authored
Co-authored-by:
-
Alexander Matveev authored
-
- 22 May, 2024 2 commits
-
-
Tyler Michael Smith authored
Pass the CUDA stream into the CUTLASS GEMMs, to avoid future issues with CUDA graphs
-
Michael Goin authored
-
- 20 May, 2024 1 commit
-
-
Alexander Matveev authored
-
- 16 May, 2024 1 commit
-
-
Tyler Michael Smith authored
-
