Repository navigation

[ROCm][Model][Bugfix] Enable GLM-5.2-MXFP4 on the deepseek_v32 path and fix sparse attention correctness - #51915

Merged

dllehr-amd merged 52 commits into

vllm-project:mainfrom

jhu960213:jhu96/enable-glm52-in-models

Sep 22, 2026

Commits on Aug 12, 2026

Added a guard to dsv_32 fused_norm_rope as ROCm Triton passes no launch_pdl kwarg as PDL is a CUDA kernel launch scheduling feature for Hopper+ archs (sm90)
jhu960213
committed
73a1c79
The batched gemm is already imported in aiter ops so changing to use wrapper instead of calling it directly, also fixed input args mismatch when calling with original batched gemm kernel
jhu960213
committed
12ed6ed
Added power of 2 guard for fp4 bmm, dropped sparse MLA block size override in the v32 path, and refactored duplicatged aiter ops module package imports along the v32 path part of rocm.py
jhu960213
committed
d104154
Seperated vendor specific logic and kept the shared common fused kernl platform agnostic, moved AMD specific shuffle indexer kv layout undr block-size > 1 routing to amd backend, extended ROCMAiter…
jhu960213
committed
d91ae1b
Parametrized model internal configs and used corrected values for both DS3.2 and GLM5.2, removed gate for cuda only since its triton backend and we gate on whether the current platform supports fp8
jhu960213
committed
70d1891
Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
825a834
Update get_supported_block_sizes() in rocm.py to prevent duplicated instantiated block sizes

jhu960213
and
jamesETsmith
authored
e1810f0
Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
df1e2a0
Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
ef4c74f
Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
9e46c41

Commits on Aug 13, 2026

Commits on Aug 20, 2026

Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
d4debcb

Commits on Aug 21, 2026

Commits on Aug 26, 2026

Commits on Aug 27, 2026

Commits on Sep 1, 2026

Commits on Sep 9, 2026

Commits on Sep 10, 2026

Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
99226d1

Commits on Sep 15, 2026

Commits on Sep 16, 2026

Commits on Sep 20, 2026

Merge branch 'main' into jhu96/enable-glm52-in-models
jhu960213
authored
e8967b3

Commits on Sep 21, 2026