tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity
2,419 Commits 319 Branches 32 Tags
f6ed8ddc01bb77ebe23f0b419939a384fd85dafd
Commit Graph

2419 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Dan Saunders
f6ed8ddc01 fix 2025-09-17 13:44:26 -04:00
Dan Saunders
556d6448fe fix 2025-09-17 13:44:26 -04:00
Dan Saunders
5c2229721d diag 2025-09-17 13:44:26 -04:00
Dan Saunders
d7de6b0e96 grouped_mm 2025-09-17 13:44:26 -04:00
Dan Saunders
3c6648678f numerics 2025-09-17 13:44:26 -04:00
Dan Saunders
5b19a1ea9c improve 2025-09-17 13:44:26 -04:00
Dan Saunders
cfefad1eea fix 2025-09-17 13:44:26 -04:00
Dan Saunders
125e7b5fe6 fast path 2025-09-17 13:44:26 -04:00
Dan Saunders
479b6144df tflops 2025-09-17 13:44:26 -04:00
Dan Saunders
68da65cba2 update 2025-09-17 13:44:26 -04:00
Dan Saunders
0d689bb421 cache, example 2025-09-17 13:44:26 -04:00
Dan Saunders
43ada1278a moe kernels init scaffold 2025-09-17 13:44:26 -04:00
Dan Saunders
4065bc14c6 Debug log, logging improvements (#3159) 
* simplify logging

* remove comment

* progress on debug.log

* add debug-level logger for file log

* simplify

* case insensitivity; 3rd party logging improvements

* simplify

* fix

* tests

* lint

* nits

* nit

* Update tests/test_utils_tee.py

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

* cleanup / comments

* fix

* oops

---------

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
 2025-09-17 13:27:03 -04:00
salman
e5c427f6de qat doc updates (#3162) [skip-ci] 2025-09-17 10:38:15 +01:00
Wing Lian
86d6ee7c05 upgrade trl and accelerate (#3161) 
* upgrade trl==0.23.0

* upgrade accelerate patch fix

* add hints when using gradient_checkpointing with DPO

* set gradient-checpointing properly
 2025-09-16 14:53:01 -04:00
Wing Lian
d4cff1b7bb improve setting of NCCL_P2P_DISABLE on runpod (#3132) [skip ci] 
* improve setting of NCCL_P2P_DISABLE on runpod

* use recs from review
 2025-09-16 14:52:45 -04:00
Wing Lian
1ef6c196f7 setup env vars for ray train for FSDP (#3130) [skip ci] 2025-09-16 14:52:29 -04:00
salman
58d67bf98d Migrate QAT API; fix axolotl quantize for QAT-ed models; add NVFP4 (#3107) 2025-09-12 10:55:50 +01:00
salman
0401a15888 SEO go brrr (#3153) [skip-ci] 2025-09-12 10:55:11 +01:00
NanoCode012
fcfc13d710 feat(doc): update thinking and chat_template notes (#3114) [skip ci] 
* feat: update thinking and chat_template notes

* fix: grammar
 2025-09-12 14:45:18 +07:00
salman
9406c0c488 log before eval step (#3148) [skip-ci] 2025-09-11 11:19:30 +01:00
Dan Saunders
1b53c49e1a text diffusion training plugin (#3067) 
* diffusion training plugin

* cleanup

* nits

* fixes + improvements

* add back in reinit_weights (clobbered?); masking / pretrain fixes

* nits

* cleanup; tests draft

* sample generation, tests fixes

* fixes

* nits

* add inference support; add auto-mask token support

* nits

* nits

* progress

* simplify logging

* lint

* prefix args with diffusion_

* coderabbito

* tests fix

* nit

* nits

* cleanup + nits

* nits

* fix SFT sample gen

* fixes

* fix

* comments

* comments

* lint

* reward model lora fix

* cleanup; fix pretraining_dataset case

* gradio inference

* update cfgs

* update cfgs

* train, generation parity, cleanup

* fix

* simplify

* test

* test fix
 2025-09-10 20:27:00 -04:00
NanoCode012
b71482cec5 Feat: add hunyuan v1 (#3016) 
* feat: add hunyuan cce support

* feat: update cce docs

* feat: add multipack support for granite and hunyuan

* feat: add hunyuan docs and example config

* feat: update readme instructions to include CCE installation

* fix: chat template log appearing despite tokenizer already having template

* feat: add vram usage

* fix: remove duplicate cce install

* fix: use latest commit of PR in case rebased/pushed

* Revert "fix: use latest commit of PR in case rebased/pushed"

This reverts commit 8b60aa00de.

* feat: update doc as upstream merged
 2025-09-10 09:03:30 +07:00
NanoCode012
79103b01ca Feat: add seedoss (#3104) [skip ci] 
* feat: add seedoss cce

* feat: add seedoss config and docs

* fix: shouldn't have target modules with target linear

* feat: add vram numbers

* fix: hf link

* fix: name

* fix: support multipack seedoss

* fix: merge error

* feat: update seedoss instructions for transformers release
 2025-09-10 09:01:02 +07:00
salman
9640338d37 Default include_tkps to true (#3134) 
* default true

* force e2e

* causal trainer only

* fix eval loggin [skip-ci]

* revert setup.py

* force tests

* guarding

* guarding

* fix test case

* use evaluate [skip-e2e]

* use evaluate [skip-e2e]

* kick off ci

* fixing

* reverting
 2025-09-09 10:50:21 -04:00
Wing Lian
b5d4c7ff54 allow 1% deviation for codecov (#3138) [skip ci] 2025-09-07 11:01:03 -04:00
Seungduk Kim
8fd9221f13 Add ipo as an rl type that shares DPODataset config (#3128) 
* Add `ipo` as an `rl` type that shares DPODataset config

* chore: lint

---------

Co-authored-by: Wing Lian <wing@axolotl.ai>
 2025-09-07 10:49:10 -04:00
github-actions[bot]
bf00f29f3a chore: update pre-commit hooks (#3137) [skip ci] 
Co-authored-by: djsaunde <1245942+djsaunde@users.noreply.github.com>
 2025-09-07 10:33:20 -04:00
NanoCode012
1d32278755 feat: upgrade transformers to v4.56.1 (#3127) 
* feat: upgrade transformers to v4.56

* fix handling of CP/SP now that position_ids are default even for unpacked sequences

* feat: monkeypatch list_repo_templates

* fix: apply patch for tests only

* see if updated main works at least

* fix: update to patch release and remove monkeypatch

* remove fsdp2 eval patch

---------

Co-authored-by: Wing Lian <wing@axolotl.ai>
 2025-09-05 11:00:54 -04:00
NanoCode012
c6ae5c43cb fix: chat template jinja file not being loaded during inference (#3112) 
* fix: chat template jinja file not being loaded during inference

* fix: bot comment
 2025-09-03 16:25:09 -04:00
yardenhoch
efa1da52d5 Center rewards coefficient (#3124) 
* feat: add center_rewards_coefficient for reward modeling

- Add center_rewards_coefficient parameter to Pydantic schema with paper reference
- Pass parameter through base builder and causal builder to training args
- Add documentation section with usage examples and theoretical background
- Enable parameter in reward modeling example configs with recommended value
- Enables reward centering for improved training stability in RLHF workflows

Implements auxiliary loss from Eisenstein et al. 2023 (https://huggingface.co/papers/2312.09244)
to incentivize mean-zero reward outputs without post-training normalization.

* Update description

* test: add unit tests for center_rewards_coefficient integration

* Update src/axolotl/core/builders/base.py

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>

* Update docs/reward_modelling.qmd

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>

* Update docs/reward_modelling.qmd

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>

* reference to TRL documentation.

* add new reward model configuration for qwen3 with comprehensive parameters

* Verified center_rewards_coefficient is correctly passed through the trainer builder to training arguments.

* Refactor reward modeling documentation to consolidate information on center_rewards_coefficient

* Remove unit tests for center_rewards_coefficient integration as part of codebase cleanup.

* linting

* nit

* Apply suggestions from code review

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>

* lint

---------

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>
Co-authored-by: Salman Mohammadi <salman.mohammadi@outlook.com>
 2025-09-03 16:22:37 -04:00
mhenrichsen
48db520d92 Create 270m-qlora.yml (#3075) [skip ci] 
Adds 270m gemma3 qlora
 2025-09-03 16:20:32 -04:00
NanoCode012
53a0c1f39c feat: add peft_trainable_token_indices (#3062) 
* feat: add peft_trainable_token_indices

* feat: add warning compat with fix_untrained_tokens
 2025-09-03 01:48:01 -04:00
github-actions[bot]
4cc6038d52 chore: update pre-commit hooks (#3122) [skip ci] 
Co-authored-by: djsaunde <1245942+djsaunde@users.noreply.github.com>
 2025-09-03 01:41:34 -04:00
NanoCode012
e48aa8a5b1 feat(doc): improve visibility for colab notebooks (#3110) [skip ci] 
* feat: improve visibility for colab notebooks

* fix: link to GH colab

* feat: change to badge and move higher
 2025-09-03 01:40:53 -04:00
xuyifann
24aba5caca Clamping the len of dataloader to minimum of 1 (#3100) [skip ci] 
* Clamping the len of dataloader to minimum of 1

* linter reformat
 2025-09-03 01:40:27 -04:00
Wing Lian
06bebcb65f run cu128-2.8.0 e2e tests on B200 (#3126) 
* run cu128-2.8.0 e2e tests on B200

* not an int 🤦

* fix yaml
 2025-09-02 13:13:23 -04:00
Dan Saunders
231a67e70b Streaming SFT support (#3101) 
* working

* fixes

* deprecate --iterable; cleanup

* pretrain_multipack_buffer_size -> streaming_multipack_buffer_size

* improvements

* tests

* remove unused

* docs, examples

* nit

* nit

* add val_set_size validation

* val

* nit

* min

* coderabbito

* cleanup

* nit

* add depr warning, cleanup

* nit

* fix test, fix quarto

* fix

* review comments

* review comments

* fix
 2025-09-02 12:08:44 -04:00
Wing Lian
0094a2d744 support for tiledmlp for GPT-OSS (#3116) 
* fix use of flex attn kwargs and add support for tiledmlp for GPT-OSS

* add logging back

* update deps
 2025-08-29 13:52:49 -04:00
Wing Lian
7ed40f1d70 automatically set env vars for single gpu deepspeed zero3 (#3118) [skip ci] 
* automatically set env vars for single gpu deepspeed zero3

* use setdefault
 2025-08-29 13:36:47 -04:00
VED
5b6ec2820f patch for ds_grads_remaining in deepspeed (#3102) [skip ci] 
* patch deepspeed

* deepspeed patch for ds_grads_remaining

* patch in Patchmanager

* chore: lint

* deepseed utils

* chore2

* patch ds_grads_remaining chore

* chore lint

* chore lint

* remove torch.nn patch

* lint

* Update src/axolotl/monkeypatch/utils.py

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>

* patched with checkpointwarapper

* lint

* only apply deepspeed patch when using activation offloading

---------

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>
Co-authored-by: Wing Lian <wing@axolotl.ai>
 2025-08-29 12:12:09 -04:00
Wing Lian
6afba3871d Add support for PyTorch 2.8.0 (#3106) 
* Add support for PyTorch 2.8.0

* loosen triton requirements

* handle torch 2.8.0 in setup.py

* fix versions

* no vllm for torch 2.8.0

* remove comment

Co-authored-by: NanoCode012 <nano@axolotl.ai>

---------

Co-authored-by: NanoCode012 <nano@axolotl.ai>
 2025-08-28 09:10:40 -04:00
Dan Saunders
dc338c3b0e Update .coderabbit.yaml (#3109) [skip ci] 
Oops, should be false.
 2025-08-27 09:50:52 -04:00
salman
d0d2fc5606 Tokens per second logging [skip-e2e] (#3072) 2025-08-27 09:10:14 +01:00
Wing Lian
e1131e9619 make always skip_move_to_device default as true (#3084) 2025-08-26 09:30:22 -04:00
Wing Lian
c4c4b90638 add tokenizer_save_jinja_files to keep legacy behavior of including chat template in tokenizer_config.json (#3093) 
* add tokenizer_save_jinja_files to keep legacy behavior of including chat template in tokenizer_config.json

* fix test import
 2025-08-26 09:30:04 -04:00
Wing Lian
0e9945e3b9 deploy training jobs to baseten w truss in axolotl cli (#3086) [skip ci] 
* deploy training jobs to baseten w truss in axolotl cli

* cleanup
 2025-08-26 09:29:50 -04:00
NanoCode012
0de254a0d0 feat: add gemma3_text attention handling for lora kernels (#3103) 2025-08-26 16:47:26 +07:00
Dan Saunders
79ddaebe9a Add ruff, remove black, isort, flake8, pylint (#3092) 
* black, isort, flake8 -> ruff

* remove unused

* add back needed import

* fix
 2025-08-23 23:37:33 -04:00
Dan Saunders
eea7a006e1 make multipack sampler patch explicit (#3096) 
* make multipack sampler patch explicit

* combining
 2025-08-22 14:29:10 -04:00