[Mamba1] - Kernel Level Chunk Alignment for Prefix Caching #34798
DarkLight1337
merged
16 commits into
vllm-project:main
from
Josephasafg:mamba1-chunk-alignment-upstream
Added chunk alignment to selective scan
b2c42e48
gemini-code-assist
commented
on 2026-02-18
Merge branch 'main' of https://github.com/vllm-project/vllm into mamb…
a7584490
Removed unused kchunksize
0cf2b77f
Removed kNThreads
52db1e52
Merge branch 'main' into mamba1-chunk-alignment-upstream
063fe4db
Merge branch 'main' into mamba1-chunk-alignment-upstream
d055445d
Removed logic from mamba1 attn and used in the mixer
8a2e34a6
Used existing computing logic both both mamba types when aligning chunks
732167b5
mypy fix
0dce84ad
Merge branch 'main' of https://github.com/vllm-project/vllm into mamb…
39390104
Merge branch 'main' into mamba1-chunk-alignment-upstream
d407054a
Added missing func arg
0ba69307
Chaned func signature to kwargs
790405ec
Moved block_idx_completed
225c0aa2
tdoublep
approved these changes
on 2026-02-27
Fixed params for mamba ssm test
f718db0f
Merge branch 'main' into mamba1-chunk-alignment-upstream
8b8d28bd
Reviewers
Assignees
No one assigned
Labels
ready
v1
Milestone
No milestone
Login to write a write a comment.
Login via GitHub