PR #46084 Fix use_cache with seq_len > 1 ( #46032)

Fix use_cache with seq_len > 1 ( #46032) #46084

vasqu merged 11 commits into huggingface:main from Ramshankar07:fix/46032-mamba2-chunked-prefill

clara2911 commented on 2026-05-20

Ramshankar07 force pushed from 6bc871f9 to 03f48bf4 69 days ago

vasqu commented on 2026-05-25

Ramshankar07 force pushed from df45189f to aac20d1c 59 days ago

vasqu commented on 2026-06-01

Ramshankar07 force pushed from 4e08cdeb to 9752bab5 57 days ago

vasqu commented on 2026-06-02

Ramshankar07 force pushed from 4e51f2fb to 720ca053 45 days ago

vasqu commented on 2026-06-15

Ramshankar07 force pushed from c769ea05 to 13efaa4c 40 days ago

vasqu approved these changes on 2026-06-22

Ramshankar07 force pushed from ffb3df62 to d8058339 35 days ago

Fix use_cache with seq_len > 1 ( #46032)

4bc25faf

Refactor convolution logic to utilize initial_states for causal convo…

d9e7fef7

Code quality

78f82289

merging the chunked-prefill elif branch, renaming

a6d446e2

making the flow cleaner

2c75d623

Refactor of has_previous_states and others.

d474fb18

adding in require_* decorator format

f48bec7e

adding in __init__

fd5dcb03

Refactor Mamba2Mixer to make it close to qwen 3 implementation and im…

8e9fa80e

Updated input handling and added separate tests for CPU and torch dev…

34241027

Ramshankar07 force pushed from e7442247 to 34241027 35 days ago

adjust tests slightly to require kernels instead

bb795032

vasqu approved these changes on 2026-06-23

vasqu merged 892be11b into main 35 days ago

Reviewers

vasqu

clara2911

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

transformers Fix use_cache with seq_len > 1 ( #46032) #46084 Merged

Fix use_cache with seq_len > 1 ( #46032) #46084

transformers
Fix use_cache with seq_len > 1 ( #46032)
#46084

Merged