Issues

• Is there an example of training LLM using praxis?

  google/praxis#87 · Filter by author knightXun opened on Dec 17, 2024

• Support for loading fp8 checkpoint

  google/praxis#68 · Filter by author wenscarl opened on May 21, 2024
  12

• Add activation checkpoint offloading

  google/praxis#63 · Filter by author jon-chuang opened on May 6, 2024

• Cross-layer attention weight sharing fails in different scopes

  google/praxis#52 · Filter by author mqyqlx opened on Mar 19, 2024
  3

• Incorrect conversion from tf dtype to jax dtype

  google/praxis#49 · Filter by author backpropper opened on Feb 2, 2024

• [Feature Request] Need Matmul Attention layer instead of Einsum to support GPU running

  google/praxis#46 · Filter by author MoFHeka opened on Jan 27, 2024

• [Feature Request] Need ZeRo-1/2 to cooperate with PP+TP+DP. Which may more faster than FSDP sometimes.

  google/praxis#45 · Filter by author MoFHeka opened on Jan 26, 2024

• gpu_fast_attention not passing segment_ids to jax pallas attention mha

  google/praxis#30 · Filter by author Cjkkkk opened on Oct 2, 2023

• Any publicly available document?

  google/praxis#7 · Filter by author Lekja00160612 opened on Mar 15, 2023
  1