google/praxis

Open All

9 open issues indexed

Any publicly available document? Open

google/praxis#7 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 16, 2026
gpu_fast_attention not passing segment_ids to jax pallas attention mha Open

google/praxis#30 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026
[Feature Request] Need ZeRo-1/2 to cooperate with PP+TP+DP. Which may more faster than FSDP sometimes. Open

google/praxis#45 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026
[Feature Request] Need Matmul Attention layer instead of Einsum to support GPU running Open

google/praxis#46 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026
Incorrect conversion from tf dtype to jax dtype Open

google/praxis#49 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 16, 2026
Cross-layer attention weight sharing fails in different scopes Open

google/praxis#52 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026
Add activation checkpoint offloading Open

google/praxis#63 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026
Support for loading fp8 checkpoint Open

google/praxis#68 · 12 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026
Is there an example of training LLM using praxis? Open

google/praxis#87 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 16, 2026

Get new issues in your inbox