AI-Hypercomputer/maxtext

Issues

• https://us-python.pkg.dev/gce-ai-infra/maxtext-build-support-packages/simple/ not public

  #758 opened 21 days ago by emergenz
  6

• Flash attention - head_dim 64

  #1047 opened a month ago by peregilk
  0

• Why logit checker has such a high tolerance?

  #1021 opened a month ago by hugoabonizio
  0

• converting Gemma maxtext compatible checkpoint to Hugging Face format

  #829 opened 4 months ago by salrowili
  4

• PGLE doesn't work for Tensor Parallelism

  #1005 opened 2 months ago by wang2yn84
  3

• Cannot do inference in float32

  #595 opened 3 months ago by borisdayma
  5

• Long Context

  #801 opened 3 months ago by peregilk
  4

• converted mlperf gpt3 ckpt starts with a worse loss

  #887 opened 3 months ago by gramesh-amd
  26

• nucleus top_p sampling seems wrong? (edit: nvm, read and tested the code wrong)

  #950 opened 3 months ago by honglu2875
  1

• Training more than one epoch

  #914 opened 3 months ago by peregilk
  4

• Mask is being ignored when cudnn_flash_attention is used

  #878 opened 3 months ago by finbarrtimbers
  2

• Support nsys profiler upload in all cases

  #911 opened 3 months ago by gobbleturk
  0

• Standalone checkpoint write seems to have memory leak

  #831 opened 3 months ago by bernardhan33
  2

• Test

  #919 opened 3 months ago by shralex
  0

• Move maxtext docker images being built to artifact registry

  #904 opened 3 months ago by parambole
  0

• Unable to recover after checkpoint saving

  #868 opened 4 months ago by peregilk
  2

• Support beam search

  #594 opened 8 months ago by borisdayma
  0

• Support for RecurrentGemma

  #605 opened 8 months ago by cyrilzakka
  0

• DEFAULT_MASK_VALUE causes gradient explosion and nan loss on deep models

  #614 opened 8 months ago by logicchains
  2

• Support LoRA training

  #609 opened 8 months ago by hxssgaa
  2

• llama_or_mistral_ckpt.py file requiring checkpoints in local file system

  #674 opened 7 months ago by shivajid
  0

• Llama3

  #683 opened 3 months ago by peregilk
  2

• Support target masking (aka loss masking or label masking) for SFT datasets

  #736 opened 6 months ago by jmschndev
  0

• How to implement 1F1B pipeline parallelism in Jax?

  #752 opened 6 months ago by MoFHeka
  1

• Inconsistent environment variable names

  #775 opened 5 months ago by gabeweisz
  0

• Multihost training collapses from time to time when loading the next batch

  #786 opened 5 months ago by YUE-FAN
  3

• Make MaxText as Python Modules

  #819 opened 5 months ago by JoeZijunZhou
  0

• Converting LLama3.1 405B checkpoint - Requesting multipass checkpoint conversion

  #864 opened 3 months ago by shivajid
  3

• Cannot see multiple GPUs when using Slurm (with proposed fix)

  #865 opened 4 months ago by gabeweisz
  0

• Inconsistent code formatting

  #735 opened 3 months ago by jmschndev
  0

• Error loading mlperf gpt3 checkpoint after pax to maxtext conversion

  #879 opened 3 months ago by gramesh-amd
  14

• mlperf gpt3 ckpt permission issues

  #847 opened 3 months ago by gramesh-amd
  11

• Cannot load the paxml gpt3 tokenizer

  #875 opened 3 months ago by gramesh-amd
  7

• How to load tfrecords from local file system for Mlperf training?

  #844 opened 4 months ago by gramesh-amd
  3

• Question: Gradient Accumulation

  #607 opened 4 months ago by thiagolaitz
  6

• FlashAttention Support - TPUv3

  #791 opened 4 months ago by maciek-pioro
  1

• aqtp release 0.8.0 breaking dependencies

  #849 opened 4 months ago by bernardhan33
  1

• Gemma 2 support

  #733 opened 5 months ago by borisdayma
  3

• `hf_access_token` only effective for loading gated datasets, not gated tokenizers

  #734 opened 5 months ago by jmschndev
  0

• Outdated links in `First_run.md`

  #776 opened 5 months ago by emergenz
  1

• Eval on C4?

  #711 opened 6 months ago by tjingrant
  1

• Update Inference Microbenchmark scripts

  #660 opened 7 months ago by jon-chuang
  0

• How to convert a model to parameter only checkpoints (unscanned) on a CPU VM

  #634 opened 8 months ago by hosseinsarshar
  2

• Reproducing pure computation TFLOPs

  #624 opened 8 months ago by prrathi
  4

• Asignación

  #622 opened 8 months ago by Cyberwoodd
  1

• Clarification: how does Llama-2-7b fit on a v4-8 when using Adam?

  #606 opened 8 months ago by rodrigo-f-nogueira
  3

• Consolidate inference related logic under jetstream-maxtext

  #612 opened 8 months ago by ahg-g
  1

• Support Qwen1.5

  #585 opened 8 months ago by Muhtasham
  1

• Gemma instructions were deleted in commit

  #579 opened 9 months ago by emergenz
  2

• Issues running test_llama2_7b.sh on TPU VM v3-8

  #572 opened 9 months ago by korney3
  1