Repository navigation
goreleaser experiements
Pre-release
Pre-release
·
459 commits
to main
since this release
01b28aa
This commit was signed with the committer’s verified signature.
The key has expired.
SamuelMarks
Samuel Marks
Changelog
- 648098c Add GRPO demo notebook: grpo_llama3_demo.ipynb
- ca75fe6 Add Gemini CLI for PR review
- 9f21bfc Add Gemini CLI for PR review
- c38e480 Add Jet as codeowner
- 87e44f3 Add Megascale Hangs Playbook as a how-to guide
- a15fc00 Add Qwen3 Maxtext to vLLM weight mapping
- 48eb50d Add
mtc_data_parallelismconfig for multi-tier checkpointing. - 17f9e48 Add
mtc_data_parallelismconfig for multi-tier checkpointing. - 6a4ff10 Add automated check for docs build
- c1eb528 Add checkpoint for "out_proj" in attention_mla path.
- 691ea12 Add codeowners for model bring-up
- 1d9ef55 Add codeowners for model bring-up
- 291ec23 Add config for Gemma3-12B on v6e-256/512/1024
- 86e99b8 Add data pipeline perf in explanations
- ef24250 Add data pipeline perf in explanations
- d0ce195 Add deepseek3 conversion utility: orbax scan to hf
- c48e6e7 Add doc: pretraining with real data
- fd20367 Add documentation for supported models and architectures in MaxText
- 61a4170 Add documentation to run SFT with Deepseek-V3 model
- 87b9a17 Add documentation to run SFT with Deepseek-V3 model
- bcd2f17 Add multimodal documentation
- ca5282b Add pure-text Qwen3-Omni configs as base
- e82887e Add script to run evaluation before and after SFT training
- 0e03921 Add support for Qwen3-Next Model Architecture
- 73d2586 Add user guide to update MaxText dependencies with seed-env
- d9cbcb2 Add workflow for unit tests with Pathways backend
- ee4b38a Added jax impl for gated full attention
- f588ed5 Adding documentation for optimized model tiers
- 3ccff8d Adding more codeowners
- 9e69232 Adding more condeowners
- 4fb383a Adding more people to approvers
- fede8e9 Adds a
flops_measuredproperty toTrainerContextand sets it toTruein MaxText's hook if TFLOPs is already calculated by MaxText - 4160978 All-in-one commit for new pw_recipe modularization
- f828172 All-in-one commit for new pw_recipe modularization
- 3a137c8 Allow user-specified num_slices
- af21663 Alphabetize ReadTheDocs explanations and references
- b68940a Alphabetize ReadTheDocs guides
- 2b25adf Avoid JAX dtype error on gpt3-6b.
- b3d8dc6 Changing config to remove 'sequence' from activation_vocab
- 35d82e6 Continue pinning Tunix dependency for nightly builds
- 8d671b4 Continue pinning Tunix dependency for nightly builds
- 420a9da Convert checkpoint for Qwen3-4B-Thinking-2507
- d2f608d Copybara import of the project:
- 8bf5cb1 Create maxtext gpu package and unit/integration tests
- c9d82e5 Disable setting specific profiler options on Pathways backend.
- 0a65506 Enable padding batch
- 9733a99 Explicitly shard input tensors across mesh devices
- 9171970 Fix GRPO & SFT links
- 77c12bd Fix Gemma3-4B e2e airflow test
- 81d17d3 Fix base_output_directory & command formatting for end_to_end & convergence test scripts
- b5fbad6 Fix bugs in uploading safetensors to GCS
- 622c0e6 Fix commands and colab link in MM readme
- 411782a Fix create_nnx_model() to load NNX checkpoint, verified with vllm_decode
- 09b5316 Fix docs build for RTD
- 5174ee6 Fix gemini-cli trigger issue
- de9de1d Fix mismatched quote in
end_to_end/tpu/test_convergence_1b_params.sh - fd2ff2b Fix token sorting with custom vjp in
moe.py - 58726c1 Fixed vllm install
- d4495e1 Fixes accuracy issues with backward pass of ring-of-experts technique
- 4b6fa6b Format code with pyink
- 3fa7c90 GPT-OSS: Add user guide and tests
- 2a83b36 GPT-OSS: Add user guide and tests
- f486fa9 Honor dataset_path when colocated_python_data_input=False.
- addd60a Implement auto-closing of build failure issues after three consecutive successes.
- 252cd4d Include venv creation in README installation instructions
- 5d71d13 Initialize pathwaysutils in sft_trainer to run SFT with pathways
- c511488 Install Tunix & Qwix from pypi
- 8748304 Introduce a batch-split version of DeepSeekV3
- 774dfd7 Llama4 vision layers NNX migration
- 75f44d7 Mainly internal tokenizer path fix
- 1c6f5a2 Mainly internal tokenizer path fix
- 05a95b7 Merge branch 'main' into docs-reorg
- e4067bf Merge branch 'main' into docs-reorg
- e8fa373 Merge branch 'main' into nicogrande/sft-llama4-tiling
- be95c02 Merge branch 'main' into nicogrande/sft-llama4-tiling
- 1bd9f00 Merge branch 'main' into nicogrande/sft-llama4-tiling
- 23d3a2e Merge branch 'main' into nicogrande/sft-llama4-tiling
- 99b2b5c Merge branch 'main' into nicogrande/sft-llama4-tiling
- 44d6528 Merge pull request AI-Hypercomputer#1974 from mesakhcienet:feat/simple-decoder-layer-nnx
- 9033e75 Merge pull request AI-Hypercomputer#1994 from SamuelMarks:auto_check_and_precommit_and_bash_code_style_sh
- 5e6a7ca Merge pull request AI-Hypercomputer#1999 from SamuelMarks:qa_MaxText
- 13872bd Merge pull request AI-Hypercomputer#2088 from AI-Hypercomputer:chengnuojin-mistral-migration
- 5fe5ca6 Merge pull request AI-Hypercomputer#2162 from AI-Hypercomputer:migrate_gemma_to_nnx
- f061537 Merge pull request AI-Hypercomputer#2166 from AI-Hypercomputer:chengnuojin-mixtral-migration
- 3b22b3d Merge pull request AI-Hypercomputer#2198 from CIeNET-International:feat/Migrate-DotProductAttention-to-NNX
- c8022db Merge pull request AI-Hypercomputer#2208 from hx89:hx/zero1_ga
- 868e0f4 Merge pull request AI-Hypercomputer#2223 from SamuelMarks:update-workflows
- e9266c8 Merge pull request AI-Hypercomputer#2242 from AI-Hypercomputer:chengnuojin-seq-tiling
- d513a4f Merge pull request AI-Hypercomputer#2256 from DannyLiCom:lidanny/pw_recipe_modularization
- ebb3df9 Merge pull request AI-Hypercomputer#2260 from AI-Hypercomputer:numpy-bottleneck
- e6aaca5 Merge pull request AI-Hypercomputer#2269 from AI-Hypercomputer:hengtaoguo-nnx-llama4
- e6792bb Merge pull request AI-Hypercomputer#2270 from AI-Hypercomputer:jax_xla_pallas
- 4ec4c72 Merge pull request AI-Hypercomputer#2271 from AI-Hypercomputer:arch_overview
- ca2ae66 Merge pull request AI-Hypercomputer#2273 from AI-Hypercomputer:gagik-supported-models
- cbd599f Merge pull request AI-Hypercomputer#2281 from AI-Hypercomputer:grpo_collab
- 613101d Merge pull request AI-Hypercomputer#2300 from AI-Hypercomputer:qinwen/add_xprof_doc
- 4673643 Merge pull request AI-Hypercomputer#2301 from AI-Hypercomputer:aireen/padding_batch
- 9784475 Merge pull request AI-Hypercomputer#2313 from AI-Hypercomputer:api_sever_v1
- 9cec359 Merge pull request AI-Hypercomputer#2326 from AI-Hypercomputer:add_owner
- 7b33d57 Merge pull request AI-Hypercomputer#2345 from melissawm:fix-docs-2
- 0a364fe Merge pull request AI-Hypercomputer#2350 from AI-Hypercomputer:aireen/gemma3-multi-image
- 84f3ad6 Merge pull request AI-Hypercomputer#2354 from bzantium:feature/AI-Hypercomputer#2344
- 9246ba9 Merge pull request AI-Hypercomputer#2355 from AI-Hypercomputer:qwen3-sft-collab
- faaf913 Merge pull request AI-Hypercomputer#2356 from melissawm:ci-docs-check
- bbe6d88 Merge pull request AI-Hypercomputer#2357 from AI-Hypercomputer:add_jet
- 73f6306 Merge pull request AI-Hypercomputer#2358 from AI-Hypercomputer:rbierneni-add-pinned-github-commits
- 8d9588a Merge pull request AI-Hypercomputer#2361 from SamuelMarks:v2-layout2
- 974323d Merge pull request AI-Hypercomputer#2367 from AI-Hypercomputer:rbierneni-fix-0.7.2
- 08d9f20 Merge pull request AI-Hypercomputer#2368 from AI-Hypercomputer:chengnuojin-fix-vmem
- 0baff00 Merge pull request AI-Hypercomputer#2369 from AI-Hypercomputer:deepseek_sharding
- f4f286c Merge pull request AI-Hypercomputer#2370 from AI-Hypercomputer:llama2_nnx
- fc5ef85 Merge pull request AI-Hypercomputer#2371 from AI-Hypercomputer:wstcliyu/pw-unit
- c84c424 Merge pull request AI-Hypercomputer#2373 from AI-Hypercomputer:qinwen/update_sharding_moe
- 0f12f3d Merge pull request AI-Hypercomputer#2376 from AI-Hypercomputer:shuningjin-doc2
- 6f9680c Merge pull request AI-Hypercomputer#2378 from AI-Hypercomputer:sft_deepseek
- e58f933 Merge pull request AI-Hypercomputer#2380 from AI-Hypercomputer:tunix_name
- 69af21b Merge pull request AI-Hypercomputer#2381 from AI-Hypercomputer:aireen/data_perf_explain
- 240a59a Merge pull request AI-Hypercomputer#2382 from AI-Hypercomputer:shuningjin-oss8
- 1089b16 Merge pull request AI-Hypercomputer#2385 from AI-Hypercomputer:bvandermoon-tunix-nightly-dep
- f1c0478 Merge pull request AI-Hypercomputer#2386 from AI-Hypercomputer:hengtaoguo-sft-validation
- fafdeaa Merge pull request AI-Hypercomputer#2391 from AI-Hypercomputer:gemini_cli2
- 7e4b1d3 Merge pull request AI-Hypercomputer#2393 from AI-Hypercomputer:fix_op_redundant
- 659e1de Merge pull request AI-Hypercomputer#2394 from AI-Hypercomputer:rbierneni-move-tunix-installation
- 8d9a1d6 Merge pull request AI-Hypercomputer#2402 from melissawm:docs-reorg
- 92e59fd Merge pull request AI-Hypercomputer#2404 from AI-Hypercomputer:tunix_pinned_commit
- 07fdfd7 Merge pull request AI-Hypercomputer#2413 from AI-Hypercomputer:nicogrande/sft-llama4-tiling
- 95542d9 Merge pull request AI-Hypercomputer#2413 from AI-Hypercomputer:nicogrande/sft-llama4-tiling
- df82415 Merge pull request AI-Hypercomputer#2414 from AI-Hypercomputer:decoding_with_maxtext_pypi
- f817224 Merge pull request AI-Hypercomputer#2416 from AI-Hypercomputer:chengnuojin-update-readme
- 134c1bb Merge pull request AI-Hypercomputer#2417 from AI-Hypercomputer:tunix_pin
- df45a16 Merge pull request AI-Hypercomputer#2419 from AI-Hypercomputer:mohit/quant_doc
- 3d082f7 Merge pull request AI-Hypercomputer#2420 from AI-Hypercomputer:moba_naive_v3
- 6ac9e04 Merge pull request AI-Hypercomputer#2421 from AI-Hypercomputer:qinwen/add_gmm_tilesize
- 903a097 Merge pull request AI-Hypercomputer#2422 from AI-Hypercomputer:fix_upload_image
- 9e12944 Merge pull request AI-Hypercomputer#2423 from AI-Hypercomputer:gemini_cli_fix
- faded41 Merge pull request AI-Hypercomputer#2424 from AI-Hypercomputer:jfacevedo_update_readme
- 8e856f1 Merge pull request AI-Hypercomputer#2428 from AI-Hypercomputer:rbierneni-qwen3-dense-checkpoint
- d110597 Merge pull request AI-Hypercomputer#2429 from SamuelMarks:setup-sh-install-deps
- 4814653 Merge pull request AI-Hypercomputer#2430 from AI-Hypercomputer:llama4_text_nnx
- a6a5f32 Merge pull request AI-Hypercomputer#2435 from AI-Hypercomputer:shuningjin-nnx-qwen
- ea80920 Merge pull request AI-Hypercomputer#2437 from AI-Hypercomputer:qinwen/add_grad_bf16
- 2ba5893 Merge pull request AI-Hypercomputer#2438 from AI-Hypercomputer:lint_fix
- d6193a5 Merge pull request AI-Hypercomputer#2439 from AI-Hypercomputer:gemma3_text_nnx
- 83cbf41 Merge pull request AI-Hypercomputer#2445 from AI-Hypercomputer:bvandermoon-pypi-release-note
- bfe0046 Merge pull request AI-Hypercomputer#2446 from CIeNET-International:maxtext/release/pr-1
- ee917cd Merge pull request AI-Hypercomputer#2448 from AI-Hypercomputer:aireen/fix_decode
- 75427da Merge pull request AI-Hypercomputer#2450 from AI-Hypercomputer:hengtaoguo-conversion
- e9568ec Merge pull request AI-Hypercomputer#2451 from AI-Hypercomputer:mohit/fix_setup
- 5da271b Merge pull request AI-Hypercomputer#2454 from AI-Hypercomputer:chengnuojin-fix-readme
- dbc7eda Merge pull request AI-Hypercomputer#2455 from AI-Hypercomputer:sft_pathways
- 57bf96b Merge pull request AI-Hypercomputer#2459 from AI-Hypercomputer:xibin/v5p_1024_config
- 1a33004 Merge pull request AI-Hypercomputer#2460 from AI-Hypercomputer:owners_change
- be918e9 Merge pull request AI-Hypercomputer#2462 from AI-Hypercomputer:pyproj_req_update
- 210e9d7 Merge pull request AI-Hypercomputer#2463 from AI-Hypercomputer:kanglan/gpu_pyproj
- 4a261bf Merge pull request AI-Hypercomputer#2466 from AI-Hypercomputer:pydantic_update
- 53bdfaa Merge pull request AI-Hypercomputer#2467 from AI-Hypercomputer:shralex_codeowner_change
- 3fdd969 Merge pull request AI-Hypercomputer#2468 from AI-Hypercomputer:hengtaoguo-cpu
- af0560c Merge pull request AI-Hypercomputer#2470 from AI-Hypercomputer:chengnuojin-shardmode
- 58dba0a Merge pull request AI-Hypercomputer#2471 from AI-Hypercomputer:nicogrande/llama4-multi-image
- 5e586bf Merge pull request AI-Hypercomputer#2472 from AI-Hypercomputer:vllm_fix
- 6b1ef88 Merge pull request AI-Hypercomputer#2474 from AI-Hypercomputer:mohit/tokamax-gmm
- 9d4fa0a Merge pull request AI-Hypercomputer#2476 from AI-Hypercomputer:anisha-docker-vllm-merge
- 32c6ca3 Merge pull request AI-Hypercomputer#2478 from AI-Hypercomputer:nnx_ckpt_restore
- cc9a196 Merge pull request AI-Hypercomputer#2482 from AI-Hypercomputer:mohit/new_resharding
- 19b9f7f Merge pull request AI-Hypercomputer#2483 from AI-Hypercomputer:hengtaoguo-dag
- 6574b4e Merge pull request AI-Hypercomputer#2485 from AI-Hypercomputer:rahulasharma/add_model_tier_doc
- 07205e2 Merge pull request AI-Hypercomputer#2486 from AI-Hypercomputer:aireen/fix_logits_xlml
- 9b5ff54 Merge pull request AI-Hypercomputer#2486 from AI-Hypercomputer:aireen/fix_logits_xlml
- 793d421 Merge pull request AI-Hypercomputer#2487 from AI-Hypercomputer:parambole/maxtext_qwen3_next_v1
- 50bafeb Merge pull request AI-Hypercomputer#2488 from AI-Hypercomputer:carlosbus/training_v6e_gemma3_12b
- 2b68435 Merge pull request AI-Hypercomputer#2490 from melissawm:add-megascale-guide
- 3530b12 Merge pull request AI-Hypercomputer#2492 from AI-Hypercomputer:chengnuojin-fix-mlp
- 3d51c99 Merge pull request AI-Hypercomputer#2493 from AI-Hypercomputer:hengtaoguo-doc
- 823a63f Merge pull request AI-Hypercomputer#2496 from AI-Hypercomputer:tests_improved
- f1fcf8c Merge pull request AI-Hypercomputer#2501 from AI-Hypercomputer:anisha-clean-path
- 33b8ac1 Merge pull request AI-Hypercomputer#2504 from AI-Hypercomputer:chengnuojin-estimator
- b959943 Merge pull request AI-Hypercomputer#2505 from abhinavgoel95:patch-3
- 3789aa1 Merge pull request AI-Hypercomputer#2507 from AI-Hypercomputer:chengnuojin-tiling
- e0b5028 Merge pull request AI-Hypercomputer#2508 from AI-Hypercomputer:aireen/llama4_logits_fix
- 57a8d52 Merge pull request AI-Hypercomputer#2511 from AI-Hypercomputer:fix_linter
- 8c57202 Merge pull request AI-Hypercomputer#2514 from SamuelMarks:fix-py-version-check-in-setup-sh
- 42cb7ed Merge pull request AI-Hypercomputer#2515 from SamuelMarks:install_maxtext_extra_deps__CACHE
- cff6edc Merge pull request AI-Hypercomputer#2516 from hx89:hx/zero1_ga_fix
- 9b2f657 Merge pull request AI-Hypercomputer#2518 from AI-Hypercomputer:mohit/train_step_fix
- 3d215cd Merge pull request AI-Hypercomputer#2519 from AI-Hypercomputer:qinwen/add_up_quantize_config
- e0b0cfc Merge pull request AI-Hypercomputer#2520 from AI-Hypercomputer:bvandermoon-installation-venv
- d2a6bdf Merge pull request AI-Hypercomputer#2523 from AI-Hypercomputer:rtd_fix
- 3527353 Merge pull request AI-Hypercomputer#2527 from AI-Hypercomputer:chengnuojin-shardmap
- 1c138f7 Merge pull request AI-Hypercomputer#2528 from AI-Hypercomputer:chengnuojin-mixtral-migration
- ff5be4a Merge pull request AI-Hypercomputer#2529 from AI-Hypercomputer:rbierneni-qwen3-next-fullattention
- dbbe011 Merge pull request AI-Hypercomputer#2531 from AI-Hypercomputer:mohit/upgrade-maxtext
- 804ff4e Merge pull request AI-Hypercomputer#2534 from AI-Hypercomputer:chengnuojin-zero1-fix
- 60b1f1b Merge pull request AI-Hypercomputer#2536 from AI-Hypercomputer:bvandermoon-version-doc
- 86937a7 Merge pull request AI-Hypercomputer#2537 from AI-Hypercomputer:aireen/grain_ckpt_scale
- 5b01873 Merge pull request AI-Hypercomputer#2538 from AI-Hypercomputer:mohit/fix_docker
- a8499dd Merge pull request AI-Hypercomputer#2539 from AI-Hypercomputer:qinwen/latest-tokamax
- 72979a3 Merge pull request AI-Hypercomputer#2540 from AI-Hypercomputer:grpo_docker_rename
- 83b3519 Merge pull request AI-Hypercomputer#2541 from SamuelMarks:setup-deps-refactor
- ac1dc93 Merge pull request AI-Hypercomputer#2545 from AI-Hypercomputer:sft_eval
- d352bc9 Merge pull request AI-Hypercomputer#2550 from AI-Hypercomputer:mohit/fix_tokamax
- 64d6d9b Merge pull request AI-Hypercomputer#2557 from AI-Hypercomputer:mohit/fix_tokamax_2
- 9d5cbba Merge pull request AI-Hypercomputer#2559 from AI-Hypercomputer:hengtaoguo-doc
- 278d2f9 Merge pull request AI-Hypercomputer#2560 from AI-Hypercomputer:config_tokamax
- b043824 Merge pull request AI-Hypercomputer#2561 from AI-Hypercomputer:hengtaoguo-vl
- aebd278 Merge pull request AI-Hypercomputer#2562 from AI-Hypercomputer:chengnuojin-move-sharding
- 499d6c8 Merge pull request AI-Hypercomputer#2564 from AI-Hypercomputer:gemini_cli_trigger
- 57e0ece Merge pull request AI-Hypercomputer#2569 from AI-Hypercomputer:bvandermoon-version-doc
- bab34e5 Merge pull request AI-Hypercomputer#2570 from AI-Hypercomputer:bvandermoon-version-doc
- 151fa9f Merge pull request AI-Hypercomputer#2572 from AI-Hypercomputer:chengnuojin-sharding-fix
- a022ce3 Merge pull request AI-Hypercomputer#2574 from AI-Hypercomputer:chengnuojin-add-tiled
- 8d6acdd Merge pull request AI-Hypercomputer#2577 from AI-Hypercomputer:bvandermoon-maxtext-installation
- ab24f1e Merge pull request AI-Hypercomputer#2585 from AI-Hypercomputer:hengtaoguo-conv
- d29bb51 Merge pull request AI-Hypercomputer#2586 from AI-Hypercomputer:multislice-rl
- 4095016 Merge pull request AI-Hypercomputer#2587 from AI-Hypercomputer:bvandermoon-setup-file-path
- 51bf31d Merge pull request AI-Hypercomputer#2588 from AI-Hypercomputer:end_to_end_fix
- 0337876 Merge pull request AI-Hypercomputer#2589 from SamuelMarks:extra-deps-location
- 9e32bf4 Merge pull request AI-Hypercomputer#2596 from AI-Hypercomputer:shuningjin-ckpt-ds3
- 2dc5ffc Merge pull request AI-Hypercomputer#2597 from AI-Hypercomputer:bvandermoon-version-doc
- 341061f Merge pull request AI-Hypercomputer#2599 from AI-Hypercomputer:nicogrande/update-codeowners
- cb136bc Merge pull request AI-Hypercomputer#2605 from AI-Hypercomputer:hengtaoguo-format
- b09e255 Migrate DotProductAttention to NNX
- 3a140b7 Migrate Gemma3DecoderLayer and Gemma3ScannableBlock to NNX
- 70f91cd Migrate GemmaDecoderLayer and Gemma2DecoderLayer to NNX
- b70e39b Migrate Llama4DecoderLayer and Llama4ScannableBlock to NNX
- 698fa04 Migrate LlamaDecoderLayer to NNX
- 9cd3c65 Migrate Qwen3 to NNX
- 220487a Minor changes to
deepseek_google_test.py - 3ab6ecf MobA: Mixture of Block Attention implementation in dot_production
- 889e94c Modify the MaxText image and the default paths for server_image and proxy_image
- 5e1b120 Move MaxText installation guide to separate ReadTheDocs page
- 38f38dd Moved tunix to setup.sh
- a818e1e Moved tunix to setup.sh
- 828597c Moving 'activation_vocab' sharding rule with single axis to the top
- e7b334a Multiple workers
- d0b0f18 Refactor Megablox GMM kernels for quantization
- ab6dcf6 Refactor: 3.1 phase of RESTRUCTURE.md, focussing on setup.sh and related Python dependency & Dockerfile files
- b4e08bd Refactor: third phase of RESTRUCTURE.md
- 66132ab Remove sharding rules for q_lora and kv_lora from base.yml
- 586a395 Removing tunix condition in JAII build
- 0c5af95 Rename docker build MODE=grpo to MODE=post-training
- bef18ba Reorganize documentation
- 2c8abbe Revert "fix setup.sh for MODE=nightly"
- 29d9d5a Reverts 07fdfd7
- 81ada4a Reverts 1a33004
- 2c9501a Set the default of
use_custom_sort_vjptoFalse - 0499a5f Sft for qwen
- 099f7af Specify maxtext[tpu] package using a requirements.txt generated by seed-env. This is a prerequisite for the gpu extra in a follow-up PR.
- da8c287 Squashed commit
- e7128bc Support grain ckpt scale up/down
- 8fdac10 Support multiple images in Gemma3
- 3b485c4 Update Latest News MaxText installation link to new location
- 699595b Update README.md
- 3de8da1 Update Tunix commit in extra_deps_from_github
- 8ffe844 Update Tunix commit in extra_deps_from_github
- a24092e Update configs for tokamax
- 2e1f410 Update demo_decoding colab to install MaxText from PyPI
- af3803c Update pinned commit for Tunix
- 9a0fa6a Update pinned commit for Tunix
- d0861f9 Update typo for pyconfig to fix the breakage on the head
- 1899214 Update typo for pyconfig to fix the breakage on the head
- e960cb2 Update workflow trigger rule
- 4567d3c Updated requirements to use pinned commits
- 27eb9a9 Use correct extra_deps_from_github.txt file path in setup.sh
- c300408 [.github/workflows/*.yml] Update all
uses:to latest tagged version ; ensure license is present on each file - 01b28aa [.goreleaser.yaml] Init
- a9c4867 [.pre-commit-config.yaml] Add same items as
code_style.sh; [.pre-commit-hooks.yaml,.editorconfig] Init ; [*] Modify code to be compliant with pre-commit hooks - 87131c1 [JAX] Remove flag jax_cpu_enable_gloo_collectives.
- 92002d6 [dependencies/dockerfiles/maxtext_{db_dependencies,dependencies,gpu_dependencies,jax_ai_image}.Dockerfile]
extra_deps_from_github.txtis now only searched for in filesystem path only… soCOPYover there - 7887afe [setup.sh] Fix Python version check to correctly lexicographically compare
- 698bf40 [setup.sh] Install extra_deps_from_github.txt unconditionally ; [.Dockerfile] Cache extra_deps_from_github.txt at same time as requirements.txt ; [.txt] Use commit hash of merged-into-main google-jetstream not PR google-jetstream
- 2e7cc0d [setup.sh] Install same dependencies as
pyproject.tomlbut in a Docker cache optimised manner whence stable; related improvements - 1a33123 [src/MaxText]
pytype+pylint+pyink+codespell - 7fc9e55 [src/MaxText]
pytype+pylint+pyink+codespell - f250b2f add estimator
- 38472b1 add explicit sharding support
- cb80139 add grad_dtype
- 36d36f3 add param sharding in grpo_trainer train_step
- b15703f add progress bar for ckpt conversion
- 68122a8 add tranposed gmm tiling for wo
- 5dc1cd7 add vocab tiling
- 7775ebf add vocab tiling
- 5a9858c add vocab tiling in readme
- ee4e8cc add weight_sum_fp32 config
- 23147c9 add zero-1 pyconfig rule
- 2fcff3f adding llama4 image tiling and basic batching for mm SFT.
- d447f53 adding llama4 image tiling and basic batching for mm SFT.
- 9d59fb9 adding llama4 tiling changes to decode.
- e57680b adding missing dim.
- d422caf adding missing params for image masks.
- acd9c90 adding multi-image support for llama4 sft and decode.
- 881a8ac arch overview
- 3eb77db benchmarks: v5p-1024 model configurations
- d223a20 change field_validator to cls instead of self becasue of pydantic version is updated to v2.x
- 08830db decode using vllm
- fa273e9 disaggregated llama3.1-70b w pathways
- f8031a0 feat(api_server): Add OpenAI-compatible API server for MaxText models
- 2b74960 feat(api_server): Add OpenAI-compatible API server for MaxText models
- 38b2940 feat(input_pipeline): Add support for chunking long sequences instead of truncation
- d2bc8f4 feat: migrate simple_layers from linen to nnx
- dbafe7e fix decode for image input
- 3d6b929 fix gather global error
- 5eed2f3 fix google-jetstream hash link
- 71e99b2 fix llama4 multimodal logits
- f7236a5 fix logits
- ea4b04d fix missing page_state
- 9167673 fix readme
- 91dcc69 fix setup.sh for MODE=nightly
- e8c3ab8 fix setup.sh for MODE=nightly
- 18fcddc fix sharding refactored functions
- 287bc39 fix tokamax
- 57fd0e4 fix tokamax splash
- b74da04 fixing _apply_embedding call.
- 55ee65c fixing decode linting.
- 3eff680 fixing merge mm embeddings.
- cbcd022 fixing pad tile normalization.
- 6589009 fusing mask + image SFT padding methods.
- f27daa6 linting fixes.
- a534a62 minor change to only do bf16 conversion when GA > 1
- ad74a01 mistral nnx migration
- b56369b mixtral nnx migration
- af9d942 new resharding API
- b8c957b parent ea80920 author Nicolas Grande nicogrande@google.com 1758826223 +0000 committer Nicolas Grande nicogrande@google.com 1759792588 +0000
- beac7ed patch zero1_ga to fix mem copy in fsdb
- d32731a path cleanups for grpo scripts
- 24fd031 pre-commit
- 0b79904 pyink linting fixes.
- c1fdce8 pylint fixes
- 3aaad8a refactor sharding components
- 4c4fbb2 refactor tiling related functions
- dbf500a remove redundant code block
- 1a4fcfb remove redundant code block
- 161f338 removing need for llama4 input changes.
- 4ba7cbd removing unecessary normalization from padding.
- c3a9001 single image tiling support for decode.
- 31b4756 skip jax 0.7.1 in github CI
- da421cf small typo in decode.
- d6f5dc8 tokamax megablox kernel
- 041ba3e update
- 7189d42 update quantization doc
- 335793b update readme to add LTXV under maxdiffusion.
- 3f3a9aa update sharding to remove extra datacopy
- 859abdf update sharding to remove extra datacopy
- cc5053f update shardmap in maxtext
- 44ef76c update test for new jax version
- e26eb32 updating codeowners.
- f2909cd upgrade maxext
- 1da6396 v0 of jax, xla, etc.
- 1b08d0b xprof user guide