google/maxtext issues and pull requests

#904 - Move maxtext docker images being built to artifact registry

Issue - State: open - Opened by parambole 5 months ago
Labels: enhancement

#903 - Adds a new end-to-end test for Mistral 7b

Pull Request - State: open - Opened by shralex 5 months ago

#902 - Fix lint errors

Pull Request - State: closed - Opened by shralex 5 months ago
Labels: pull ready

#901 - Refactoring Maxtext build process with stable stack

Pull Request - State: open - Opened by parambole 5 months ago

#900 - Disable zarr3 when using single controller runtime

Pull Request - State: open - Opened by shauryagup 5 months ago

#899 - Fix lint errors.

Pull Request - State: closed - Opened by shralex 5 months ago
Labels: pull ready

#898 - Add configs for Attention Block Size tuning

Pull Request - State: closed - Opened by Obliviour 5 months ago
Labels: pull ready

#897 - Maxtext Offline serverless inference code

Pull Request - State: open - Opened by vipannalla 5 months ago - 1 comment

#896 - [WIP] partial nnx impl

Pull Request - State: open - Opened by rdyro 5 months ago

#895 - Initialize jax distributed when checkpointing is enabled

Pull Request - State: open - Opened by jonb377 5 months ago - 4 comments

#894 - Add Llama 2 70B config on v5p

Pull Request - State: closed - Opened by raymondzouu 5 months ago
Labels: pull ready

#893 - Run code-style on changed files in pre-commit.

Pull Request - State: closed - Opened by shralex 5 months ago
Labels: pull ready

#892 - Run code-style on changed files in pre-commit

Pull Request - State: closed - Opened by shralex 5 months ago

#891 - Add precision option

Pull Request - State: closed - Opened by RissyRan 5 months ago
Labels: pull ready

#890 - Docker prune in all github actions

Pull Request - State: open - Opened by khatwanimohit 5 months ago - 1 comment

#889 - Add GPT-3 175B v5p MLPerf 4.0 scripts

Pull Request - State: closed - Opened by anfals 5 months ago
Labels: pull ready

#888 - convert maxtext trained orbax checkpoint to HF checkpoint

Pull Request - State: closed - Opened by jwyang-google 5 months ago - 1 comment
Labels: pull ready

#887 - converted mlperf gpt3 ckpt starts with a worse loss

Issue - State: open - Opened by gramesh-amd 5 months ago - 13 comments

#886 - stage first axes mesh

Pull Request - State: open - Opened by gobbleturk 5 months ago

#885 - Move pylint satements to the top of the file to conform with Google s…

Pull Request - State: closed - Opened by shralex 5 months ago
Labels: pull ready

#884 - Disable AOT activation offload test

Pull Request - State: closed - Opened by gobbleturk 5 months ago
Labels: pull ready

#883 - removing tensorflow_text for aarch64 compatiblity

Pull Request - State: open - Opened by rdyro 5 months ago - 1 comment
Labels: pull ready

#882 - Support older checkpoint deletion and customized checkpoint sizes

Pull Request - State: closed - Opened by bernardhan33 5 months ago

#881 - Add gpt-3 175B script for trillium

Pull Request - State: closed - Opened by raymondzouu 5 months ago
Labels: pull ready

#880 - Switch Expert axis to avoid unnecessary copy for layout change

Pull Request - State: closed - Opened by ZhiyuLi-goog 5 months ago - 2 comments
Labels: pull ready

#879 - Error loading mlperf gpt3 checkpoint after pax to maxtext conversion

Issue - State: closed - Opened by gramesh-amd 5 months ago - 14 comments

#878 - Mask is being ignored when cudnn_flash_attention is used

Issue - State: open - Opened by finbarrtimbers 5 months ago
Labels: bug, good first issue

#877 - remove attention type from gemma2 model configs

Pull Request - State: closed - Opened by wenxindongwork 5 months ago - 2 comments
Labels: pull ready

#876 - Add eval to convergence test and log metrics

Pull Request - State: closed - Opened by aireenmei 5 months ago
Labels: pull ready

#875 - Cannot load the paxml gpt3 tokenizer

Issue - State: closed - Opened by gramesh-amd 5 months ago - 7 comments

#874 - Additional Step to clean older docker images

Pull Request - State: closed - Opened by parambole 5 months ago
Labels: pull ready

#873 - [MoE][int8] add quantization to MoE dropped implementation

Pull Request - State: closed - Opened by ZhiyuLi-goog 5 months ago
Labels: pull ready

#872 - Fix mixtral8x7b model decoder references

Pull Request - State: closed - Opened by kyle-google 5 months ago - 1 comment

#871 - Add block_until_ready operation before checkpoint saving operation.

Pull Request - State: closed - Opened by abhinavclemson 5 months ago
Labels: pull ready

#870 - clarify that Flax checkpoints are expected for Gemma

Pull Request - State: closed - Opened by nhira 5 months ago
Labels: pull ready

#869 - Enable expert parallelism for dropping strategy

Pull Request - State: closed - Opened by RissyRan 6 months ago
Labels: pull ready

#868 - Unable to recover after checkpoint saving

Issue - State: open - Opened by peregilk 6 months ago - 2 comments
Labels: bug

#867 - Make running preflight optional in model scripts

Pull Request - State: closed - Opened by raymondzouu 6 months ago
Labels: pull ready

#866 - test code to produce Lab Notes - 2024-09-07.ipynb

Pull Request - State: open - Opened by bernardhan33 6 months ago

#865 - Cannot see multiple GPUs when using Slurm (with proposed fix)

Issue - State: open - Opened by gabeweisz 6 months ago
Labels: good first issue, feature request

#864 - Converting LLama3.1 405B checkpoint - Requesting multipass checkpoint conversion

Issue - State: closed - Opened by shivajid 6 months ago - 3 comments

#863 - Add MaxText run name to TensorBoard file directory

Pull Request - State: closed - Opened by bvandermoon 6 months ago
Labels: pull ready

#862 - Improve tfds perf in multihost env

Pull Request - State: closed - Opened by aireenmei 6 months ago
Labels: pull ready

#861 - Fix circ storage check for delayed case

Pull Request - State: closed - Opened by gobbleturk 6 months ago
Labels: pull ready

#860 - Add load balance loss

Pull Request - State: closed - Opened by RissyRan 6 months ago
Labels: pull ready

#859 - RA update works for all axes orders

Pull Request - State: closed - Opened by patemotter 6 months ago
Labels: pull ready

#858 - Add simple MLP decoder block

Pull Request - State: closed - Opened by gobbleturk 6 months ago
Labels: pull ready

#857 - Delay Activation Forwarding

Pull Request - State: closed - Opened by gobbleturk 6 months ago - 1 comment
Labels: pull ready

#856 - added run_name_prefix to tensorboard

Pull Request - State: closed - Opened by kyle-google 6 months ago - 1 comment

#855 - Temporarily pin google-cloud-aiplatform to 1.61.0

Pull Request - State: closed - Opened by bvandermoon 6 months ago
Labels: pull ready

#854 - [DRAFT] Add In Memory Changes for Pathways

Pull Request - State: open - Opened by SujeethJinesh 6 months ago

#853 - Fix kernel imports

Pull Request - State: closed - Opened by gobbleturk 6 months ago
Labels: pull ready

#852 - Add node attributes to the training benchmark

Pull Request - State: closed - Opened by bernardhan33 6 months ago

#851 - Fix kernel imports

Pull Request - State: closed - Opened by gobbleturk 6 months ago - 1 comment
Labels: pull ready

#850 - Add node attributes; Fix GCS upload; Add checkpointID to checkpointing workload

Pull Request - State: closed - Opened by bernardhan33 6 months ago - 1 comment

#849 - aqtp release 0.8.0 breaking dependencies

Issue - State: closed - Opened by bernardhan33 6 months ago - 1 comment

#848 - documenting XLA flags used by MaxText

Pull Request - State: closed - Opened by nhira 6 months ago - 1 comment
Labels: pull ready

#847 - mlperf gpt3 ckpt permission issues

Issue - State: closed - Opened by gramesh-amd 6 months ago - 11 comments

#846 - Add Llama2 config for v5p

Pull Request - State: closed - Opened by raymondzouu 6 months ago
Labels: pull ready

#845 - Adding Mixtral-8x22b

Pull Request - State: closed - Opened by rdyro 6 months ago - 2 comments
Labels: pull ready

#844 - How to load tfrecords from local file system for Mlperf training?

Issue - State: closed - Opened by gramesh-amd 6 months ago - 3 comments

#843 - Add Gemma2-27b

Pull Request - State: closed - Opened by ZhaoyueCheng 6 months ago
Labels: pull ready

#842 - Optimize overhead right before the first train_step

Pull Request - State: closed - Opened by ZhiyuLi-goog 6 months ago

#841 - Add dispatch and combine masks for dropping

Pull Request - State: closed - Opened by RissyRan 6 months ago - 1 comment
Labels: pull ready

#840 - Mlperf/4.1 grain

Pull Request - State: open - Opened by aireenmei 6 months ago - 1 comment

#838 - Llama3.1 (8B,70B) 🦙

Pull Request - State: open - Opened by khatwanimohit 6 months ago - 3 comments
Labels: pull ready

#837 - script to convert llama, mistral, mixtral checkpoints to huggingface format

Pull Request - State: closed - Opened by jwyang-google 6 months ago - 2 comments

#835 - Adds ragged attention.

Pull Request - State: closed - Opened by patemotter 6 months ago
Labels: pull ready

#834 - Integrate Badput monitoring with MaxText

Pull Request - State: closed - Opened by dipannita08 6 months ago
Labels: pull ready

#831 - Standalone checkpoint write seems to have memory leak

Issue - State: open - Opened by bernardhan33 6 months ago - 1 comment

#829 - converting Gemma maxtext compatible checkpoint to Hugging Face format

Issue - State: open - Opened by salrowili 6 months ago - 3 comments
Labels: feature request

#819 - Make MaxText as Python Modules

Issue - State: open - Opened by JoeZijunZhou 6 months ago
Labels: feature request

#817 - documenting XLA flags used by MaxText

Pull Request - State: closed - Opened by nhira 7 months ago - 1 comment

#811 - flash attention sweep

Pull Request - State: closed - Opened by Obliviour 7 months ago - 1 comment

#803 - Adding Mixtral-8x22b

Pull Request - State: closed - Opened by rdyro 7 months ago - 6 comments

#801 - Long Context

Issue - State: open - Opened by peregilk 7 months ago - 2 comments
Labels: feature request

#791 - FlashAttention Support - TPUv3

Issue - State: closed - Opened by maciek-pioro 7 months ago - 1 comment

#786 - Multihost training collapses from time to time when loading the next batch

Issue - State: open - Opened by YUE-FAN 7 months ago - 3 comments
Labels: bug

#782 - [DON'T MERGE] GCS Checkpointing Testing Workload modification

Pull Request - State: open - Opened by bernardhan33 7 months ago

#775 - Inconsistent environment variable names

Issue - State: open - Opened by gabeweisz 7 months ago
Labels: feature request

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

Issue - State: open - Opened by emergenz 8 months ago - 6 comments
Labels: bug, GPU

#752 - How to implement 1F1B pipeline parallelism in Jax?

Issue - State: open - Opened by MoFHeka 8 months ago - 1 comment
Labels: feature request

#746 - Add convergence tests on A3 GPU

Pull Request - State: closed - Opened by michelle-yooh 8 months ago
Labels: pull ready

#744 - [DON'T MERGE] GCS Distributed Training Benchmark Infra + File-parallelism + Range-read Parquet files

Pull Request - State: open - Opened by bernardhan33 8 months ago

#736 - Support target masking (aka loss masking or label masking) for SFT datasets

Issue - State: open - Opened by jmschndev 8 months ago
Labels: feature request

#735 - Inconsistent code formatting

Issue - State: closed - Opened by jmschndev 8 months ago

#683 - Llama3

Issue - State: closed - Opened by peregilk 9 months ago - 2 comments

#674 - llama_or_mistral_ckpt.py file requiring checkpoints in local file system

Issue - State: open - Opened by shivajid 9 months ago
Labels: feature request

#635 - Replace deprecated np.product with np.prod

Pull Request - State: open - Opened by gobbleturk 10 months ago

#634 - Conversion fails when JAX_COORDINATOR_ADDRESS is None

Issue - State: open - Opened by hosseinsarshar 10 months ago

#633 - Streamlined setup.sh to have fewer apt install calls and avoid purple screen of death

Pull Request - State: closed - Opened by rwitten 10 months ago
Labels: pull ready

#632 - Print Time More Accurately In MaxText

Pull Request - State: open - Opened by rwitten 10 months ago

#631 - Enable entropy on multihost CPUs.

Pull Request - State: closed - Opened by RoshaniN 10 months ago - 2 comments
Labels: pull ready

#630 - Add tests to GPU runner

Pull Request - State: open - Opened by michelle-yooh 10 months ago - 1 comment
Labels: pull ready

#629 - Fix test_tokenize unit test

Pull Request - State: closed - Opened by khatwanimohit 10 months ago
Labels: pull ready

#628 - loosen tolerance in assert_params_sufficiently_sharded

Pull Request - State: closed - Opened by ZhiyuLi-goog 10 months ago
Labels: pull ready

#627 - Add more tests for Mixtral

Pull Request - State: open - Opened by RissyRan 10 months ago

#626 - Update constraints to the latest stable

Pull Request - State: open - Opened by chajath 10 months ago

#625 - add debug functionality for per chip sizes and bytes

Pull Request - State: open - Opened by morgandu 10 months ago

#624 - Reproducing pure computation TFLOPs

Issue - State: closed - Opened by prrathi 10 months ago - 4 comments

GitHub / google/maxtext issues and pull requests