Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Sep 8, 2023

• Ensure vertex ai tensorboard experiment names are < 128 chars.

  

  Vivek Rathod

  committed
  4dfdf79

  Copy full SHA for 4dfdf79

Commits on Sep 6, 2023

• Add param conversion for CLIPVisionModelWithProjection (#53)

  wwu137

  authored
  5da278d

  Copy full SHA for 5da278d

• Supports stack-of-stacks and repeat-of-stacks in transformer decoding. (#58)

  

  Mark Lee

  authored
  cebce4b

  Copy full SHA for cebce4b

• [gcp]: create TPU with specificed service account (#61)

  Show description for 6314063
  

  zhzhyi

  and

  Zhaoyi Zhang

  authored
  6314063

  Copy full SHA for 6314063

Commits on Sep 2, 2023

• Add bastion architecture diagram in docs/ (#57)

  Show description for ad7bcef
  ya5ut

  authored
  ad7bcef

  Copy full SHA for ad7bcef

• Supports retaining intermediate output collections across scan. (#56)

  

  Mark Lee

  authored
  79de3bb

  Copy full SHA for 79de3bb

Commits on Sep 1, 2023

• Adds axlearn gcp cli. (#54)

  

  Mark Lee

  authored
  f9663da

  Copy full SHA for f9663da

• Removes timestamp information in vprint (#51)

  Show description for e24fa8c
  ruomingp

  authored
  e24fa8c

  Copy full SHA for e24fa8c

Commits on Aug 31, 2023

• [gcp]: add service account impersonation (#55)

  Show description for d11ac39
  

  zhzhyi

  and

  Zhaoyi Zhang

  authored
  d11ac39

  Copy full SHA for d11ac39

• checkpointing policy at last iteration (#52)

  Show description for a50e51f
  

  adambielski

  and

  Adam Bielski

  authored
  a50e51f

  Copy full SHA for a50e51f

Commits on Aug 28, 2023

• Adds read_state_spec to checkpointer.py. (#49)

  ruomingp

  authored
  b52fece

  Copy full SHA for b52fece

• Lists supported bundler types when reporting the unknown bundler type error (#50)

  ruomingp

  authored
  ae9805e

  Copy full SHA for ae9805e

Commits on Aug 27, 2023

• Adds Module.vprint, which uses jax.debug.print to print tensor contents (#48)

  Show description for 5c74c6c
  ruomingp

  authored
  5c74c6c

  Copy full SHA for 5c74c6c

• fix alibi logits bias order (#37)

  gyin94

  authored
  f629d82

  Copy full SHA for f629d82

Commits on Aug 26, 2023

• Remove FrozenDict. (#47)

  tuzhucheng

  authored
  0e038c6

  Copy full SHA for 0e038c6

• Adds support for generics in configs. (#44)

  

  Mark Lee

  authored
  5362123

  Copy full SHA for 5362123

Commits on Aug 25, 2023

• Removes unused DummyDecodingModel. (#46)

  

  Mark Lee

  authored
  82fa510

  Copy full SHA for 82fa510

Commits on Aug 24, 2023

• Adds support for beam search prefilling. (#42)

  

  Mark Lee

  authored
  5fd2d80

  Copy full SHA for 5fd2d80

Commits on Aug 23, 2023

• [gcp]: migrate google oauth2 credentials to google auth credentials (#43)

  Show description for da65666
  

  zhzhyi

  and

  Zhaoyi Zhang

  authored
  da65666

  Copy full SHA for da65666

• Make logits partition specs configurable in decoder. (#41)

  Show description for 4c63214
  

  xianzhidu

  and

  Xianzhi Du

  authored
  4c63214

  Copy full SHA for 4c63214

Commits on Aug 22, 2023

• Adds the axlearn.cloud module. (#40)

  

  Mark Lee

  authored
  b20659e

  Copy full SHA for b20659e

• Removes _auto_child_context_method_names. (#39)

  

  Mark Lee

  authored
  28721cc

  Copy full SHA for 28721cc

Commits on Aug 20, 2023

• Supports prefix padding with sample_decode prefill. (#38)

  

  Mark Lee

  authored
  866d031

  Copy full SHA for 866d031

Commits on Aug 17, 2023

• Fixes lion optimizer dtypes. (#36)

  

  Mark Lee

  authored
  5668213

  Copy full SHA for 5668213

Commits on Aug 13, 2023

• torch: attention softmax as float32 for bf16 (#20)

  gyin94

  authored
  09556ff

  Copy full SHA for 09556ff

Commits on Aug 10, 2023

• Revert "fix l2_norm (#21)" (#34)

  Show description for 12817d2
  

  Mark Lee

  authored
  12817d2

  Copy full SHA for 12817d2

Commits on Aug 8, 2023

• Adds initial GCP utilities. (#33)

  

  Mark Lee

  authored
  f419b8a

  Copy full SHA for f419b8a

• fix l2_norm (#21)

  gokul-uf

  authored
  af2b160

  Copy full SHA for af2b160

Commits on Aug 7, 2023

• Adds T5-UL2 configuration (#32)

  Show description for f6f69f2
  mehrnaztavan

  authored
  f6f69f2

  Copy full SHA for f6f69f2

• Allow alpha=0 in lora. (#28)

  Show description for a10f95b
  JianyuWangV

  authored
  a10f95b

  Copy full SHA for a10f95b

• Supports configuring causal=True optimization in FlashAttention mha. (#31)

  

  Mark Lee

  authored
  c3a7f7b

  Copy full SHA for c3a7f7b

Commits on Aug 6, 2023

• Add comments for the StackedTransformerLayer layer config. (#29)

  Show description for e32331e
  

  xianzhidu

  and

  Xianzhi Du

  authored
  e32331e

  Copy full SHA for e32331e

• Provides a brief overview of AXLearn in README.md (#30)

  Show description for 4eacb5e
  ruomingp

  authored
  4eacb5e

  Copy full SHA for 4eacb5e

Commits on Aug 4, 2023

• Enable building StackedTransformer layer with a sequence of layer configs. (#27)

  Show description for 8e79a1e
  

  xianzhidu

  and

  Xianzhi Du

  authored
  8e79a1e

  Copy full SHA for 8e79a1e

• Adds a flash attention layer. (#26)

  

  Mark Lee

  authored
  7e45e86

  Copy full SHA for 7e45e86