| .. |
|
Dockerfile
|
848772a3e7
glm53-flash-dflash2: target-side EAGLE3 port + rigorous KV-page blocker writeup
|
1 месяц назад |
|
UPSTREAM.md
|
1c1d3455f8
glm53-flash: NVFP4-4p67 + DFlash2-MXFP8 recipes (model cards followed, GB10-adapted)
|
1 месяц назад |
|
build.sh
|
3fa8ac759b
builds: persistent per-node buildkit cache (buildx --cache-from/to, mode=max) so version-independent apt/torch/CUDA stages are reused across recipes
|
1 месяц назад |
|
dflash.py
|
1c1d3455f8
glm53-flash: NVFP4-4p67 + DFlash2-MXFP8 recipes (model cards followed, GB10-adapted)
|
1 месяц назад |
|
patch_eagle3.py
|
848772a3e7
glm53-flash-dflash2: target-side EAGLE3 port + rigorous KV-page blocker writeup
|
1 месяц назад |
|
qwen3_dflash.py
|
1c1d3455f8
glm53-flash: NVFP4-4p67 + DFlash2-MXFP8 recipes (model cards followed, GB10-adapted)
|
1 месяц назад |
|
qwen3_dflash2.py
|
1c1d3455f8
glm53-flash: NVFP4-4p67 + DFlash2-MXFP8 recipes (model cards followed, GB10-adapted)
|
1 месяц назад |
|
versions.env
|
1c1d3455f8
glm53-flash: NVFP4-4p67 + DFlash2-MXFP8 recipes (model cards followed, GB10-adapted)
|
1 месяц назад |
Alexander Korolev
3fa8ac759b
builds: persistent per-node buildkit cache (buildx --cache-from/to, mode=max) so version-independent apt/torch/CUDA stages are reused across recipes