| FROM lmsysorg/sglang:glm-5.3-flash-arm64 | |
| # PR 36708: DFLASH mHC capture adapter | |
| COPY glm5_next.py /sgl-workspace/sglang/python/sglang/srt/models/glm5_next.py | |
| # GB10 smem retune per LibertAIDAI card: block_I=32, num_stages=1, threads=128 (169984B -> fits 101376B) | |
| COPY tilelang_kernel.py /sgl-workspace/sglang/python/sglang/kernels/ops/attention/dsa/tilelang_kernel.py | |