-
Notifications
You must be signed in to change notification settings - Fork 660
Pull requests: google/gemma.cpp
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Use the current SIMD target's width for the transposed KV cache
#1026
opened Sep 11, 2026 by
Magnushst
Loading…
Avoids repeated heap allocations/deallocations of
layout.total_bytes per attention subtask and kv_out_mem per token step by storing worker_workspaces and kv_out_mem on AttentionActivations / AttentionActivationsPtrs. This eliminates ~20M minor page faults and improves decode throughput by +18% to +26% on AMD Turin.
#1025
opened Sep 11, 2026 by
copybara-service
Bot
Loading…
Fix tiled KV cache allocation and support prefix cloning.
#1015
opened Sep 4, 2026 by
copybara-service
Bot
Loading…
Add Highway SIMD logit masking kernel for constrained decoding.
#1014
opened Sep 4, 2026 by
copybara-service
Bot
Loading…
Fix crash/abort in matmul when partition size exceeds kMaxNC.
#958
opened Jul 19, 2026 by
copybara-service
Bot
Loading…
Add direct HuggingFace safetensors loader for Gemma 4 (E2B/E4B)
#919
opened May 21, 2026 by
ssfdre38
Loading…
Add configurables for norm/rope/activation/scale/residual connection.
#287
opened Jul 3, 2024 by
copybara-service
Bot
Loading…
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.