[CPU] Enable Granite 4 / Mamba models on CPU backend #39157

Draft

Akashcodes732 wants to merge 20 commits into

vllm-project:mainfrom

Akashcodes732:feat/granite4_enablement

Commits on Apr 7, 2026

feat: initial Granite4 enablement modifications

Akash Kaothalkar
authored and
Akash Kaothalkar
committed
granite4 running on power

Akash Kaothalkar
authored and
Akash Kaothalkar
committed
granite4 cpu fallback support

Akash Kaothalkar
authored and
Akash Kaothalkar
committed
fix torch import
Akash Kaothalkar
committed
fix CustomOp Register
Akash Kaothalkar
committed
fix kwargs
Akash Kaothalkar
committed
fix: address review comments
Akash Kaothalkar
committed
fix precommit
Akash Kaothalkar
committed
Address pre-commit issues
Akash kaothalkar
committed

Commits on Apr 25, 2026

Add mamba files

Akash Kaothalkar
authored and
Akash Kaothalkar
committed
Merge remote-tracking branch 'vllm/main' into feat/granite4_enablement

Akash Kaothalkar
authored and
Akash Kaothalkar
committed

Commits on May 5, 2026

Merge remote-tracking branch 'vllm/main' into feat/granite4_enablement
Akash kaothalkar
committed
fix utils.py
Akash kaothalkar
committed
fix
Akash kaothalkar
committed
fix mamba kernels
Akash kaothalkar
committed
use cpu_types
Akash kaothalkar
committed
optimize float loading
Akash kaothalkar
committed

Commits on May 6, 2026

fix bf16 issues
Akash kaothalkar
committed
fix bf16 to fp32
Akash kaothalkar
committed
fix bf16 to fp32 .
Akash kaothalkar
committed