Spaces:

Duplicated from mrfakename/minimax-h3-ultra-fast

mayafree
/

minimax-h3-ultra-fast

Running on Zero

App Files Files Community

minimax-h3-ultra-fast / README.md

Commit History

Document complete optimization stack

28abc33

mrfakename commited on Aug 4

Gate Sol-Attn to long sequences

6f1b8ed

mrfakename commited on Aug 4

Use portable Sol-Attn backend

dd7a8ed

mrfakename commited on Aug 4

Integrate NVIDIA H3 Sol acceleration

d13df9d

mrfakename commited on Aug 4

Revert "Test PyTorch 2.13 CUDA runtime"

73feb1e

mrfakename commited on Aug 4

Test PyTorch 2.13 CUDA runtime

afa973d

mrfakename commited on Aug 4

Revert "Bank exact AdaLN curve projections"

062859f

mrfakename commited on Aug 4

Bank exact AdaLN curve projections

f0b0ab5

mrfakename commited on Aug 4

Remove lossless denoiser launch overhead

5439432

mrfakename commited on Aug 4

Avoid Triton cold-start overhead by default

997a0a3

mrfakename commited on Aug 4

Fuse exact AdaLN CUDA operations

d744aea

mrfakename commited on Aug 4

Restore quality-first 28-step default

4281716

mrfakename commited on Aug 4

Add ultra-fast residual forecasting mode

7bba4bf

mrfakename commited on Aug 4

Localize the H3 conditioner

a4e1031

mrfakename commited on Aug 4

Add adaptive H3 denoising acceleration

eb09060

mrfakename commited on Aug 4

Add optimized MiniMax-H3 NVFP4 Space

bf0e5ad

mrfakename commited on Aug 4

initial commit

bdb2ca7
verified

mrfakename commited on Jul 1