Wing Lian
1a7f048c6b
add SOAP optimizer
2025-03-31 08:33:19 -04:00
Wing Lian
76d26366ad
upstream updates for momentum change
2025-03-31 08:33:19 -04:00
Wing Lian
64fe284765
add soap optimize
2025-03-31 08:33:19 -04:00
NanoCode012
cf0c79d52e
fix: minor patches for multimodal ( #2441 )
...
* fix: update chat_template
* fix: handle gemma3 showing a lot of no content for turn 0
* fix: remove unknown config from examples
* fix: test
* fix: temporary disable gemma2 test
* fix: stop overwriting config.text_config unnecessarily
* fix: handling of set cache to the text_config section
* feat: add liger gemma support and bump liger to 0.5.5
* fix: add double use_cache setting
* fix: add support for final_logit_softcap in CCE for gemma2/3
* fix: set use_cache before model load
* feat: add missing layernorm override
* fix: handle gemma3 rmsnorm
* fix: use wrapper to pass dim as hidden_size
* fix: change dim to positional
* fix: patch with wrong mlp
* chore: refactor use_cache handling
* fix import issues
* fix tests.e2e.utils import
---------
Co-authored-by: Wing Lian <wing@axolotl.ai >
2025-03-31 13:40:12 +07:00
Wing Lian
4ba80a0e5a
fix streaming packing test ( #2454 )
...
* fix streaming packing test
* constrain amount of text generated
2025-03-29 08:30:06 -04:00
Wing Lian
c49682132b
use offline for precached stream dataset ( #2453 )
2025-03-28 23:39:09 -04:00
Wing Lian
e46239f8d3
bump liger to 0.5.5 ( #2448 )
2025-03-28 19:21:03 -04:00
Wing Lian
05f03b541a
hf offline decorator for tests to workaround rate limits ( #2452 ) [skip ci]
...
* hf offline decorator for tests to workaround rate limits
* fail quicker so we can see logs
* try new cache name
* limit files downloaded
* phi mini predownload
* offline decorator for phi tokenizer
* handle meta llama 8b offline too
* make sure to return fixtures if they are wrapped too
* more fixes
* more things offline
* more offline things
* fix the env var
* fix the model name
* handle gemma also
* force reload of modules to recheck offline status
* prefetch mistral too
* use reset_sessions so hub picks up offline mode
* more fixes
* rename so it doesn't seem like a context manager
* fix backoff
* switch out tinyshakespeare dataset since it runs a py script to fetch data and doesn't work offline
* include additional dataset
* more fixes
* more fixes
* replace tiny shakespeaere dataset
* skip some tests for now
* use more robust check using snapshot download to determine if a dataset name is on the hub
* typo for skip reason
* use local_files_only
* more fixtures
* remove local only
* use tiny shakespeare as pretrain dataset and streaming can't be offline even if precached
* make sure fixtures aren't offline
improve the offline reset
try bumping version of datasets
reorder reloading and setting
prime a new cache
run the tests now with fresh cache
try with a static cache
* now run all the ci again with hopefully a correct cache
* skip wonky tests for now
* skip wonky tests for now
* handle offline mode for model card creation
2025-03-28 19:20:46 -04:00
Wing Lian
a4e430e7c4
add override of upstream fix for multi-gpu orpo ( #2440 )
...
* add override of upstream fix
* override batch loss metrics for CPO/Simpo as well
2025-03-26 18:14:59 -04:00
Wing Lian
6cdcb8ddd5
Set the pytorch_cuda_alloc_conf env in the train module ( #2447 )
2025-03-26 18:14:43 -04:00
NanoCode012
a7811ad4a0
fix(doc): document config required to run eval_causal_lm_metrics ( #2445 ) [skip ci]
2025-03-26 18:14:29 -04:00
NanoCode012
e2da821e67
chore: minor optim changes (add apollo, improve docs, remove lion-pytorch) ( #2444 )
...
* feat: add apollo-torch
* chore: update optimizer list
* fix: deleted accidental requirements file
* fix: remove mention of deprecated lion_pytorch
2025-03-26 18:14:07 -04:00
NanoCode012
2c34a4634e
feat: add CCE for gemma3, cohere, and cohere2 ( #2443 )
...
* feat: add CCE for gemma3 and cohere1/2
* fix: change from relative import to absolute
* feat: add multipack for cohere&cohere2
* chore: improve comments
* fix: add gemma3_text
* feat: add cohere2 example
* fix: cohere forward
* fix: patch for cohere2
* feat: add command r v01 qlora sample
* chore: lint
* feat: upgrade gemma3 and gemma2 patch to use logits_to_keep
* chore: lint
* fix: add deprecate_kwarg decorator
* fix: add cce for gemma3 conditionalgeneration
* fix: gemma3 patch to defer logits calculation
* fix: patch gemma3 if given as model
* fix: remove not working config
* fix: update comments to clarify changes
* feat(doc): add supported models to readme
* fix: address difference in our cohere patch
* feat: add mistral3
* feat: add gemma
* feat(doc): update README to include gemma and mistral3 in supported models
* fix: gemma patch
* fix: import
* fix: gemma patch to be standalone
* fix: gemma3 warn about not support final_logit_softcapping
* feat: add mllama CCE
* chore: add abbireviation to doc
* fix: remove unneeded gemma3 eager warning
* fix: save processor if available
* fix: enable save processor on merge
* fix: wrong env meaning
2025-03-26 18:13:51 -04:00
NanoCode012
a9b0733f2c
Feat: Rework multimodal support (mllama, llava, pixtral, qwen2, qwen25, gemma3, mistral3) ( #2435 )
2025-03-23 11:08:51 -04:00
NanoCode012
9f00465a5c
Feat: Add support for gemma3_text and add e2e for gemma2 ( #2406 )
2025-03-22 20:33:21 -04:00
Dan Saunders
86bac48d14
cleanup for failing test ( #2436 )
2025-03-22 17:53:29 -04:00
Dan Saunders
e44953d50c
installing axolotl prior to quartodoc build ( #2434 )
...
* installing axolotl prior to quartodoc build
* simplify by installing no deps
---------
Co-authored-by: Dan Saunders <dan@axolotl.ai >
2025-03-21 13:28:13 -04:00
Dan Saunders
23f0c51d88
Sequence parallelism ( #2412 )
...
* adding easy_context as integration for now
* progress on ring attn impl
* progress on ring attn impl
* cleanup
* remove errant file
* fix req
* removing unused code
* updates
* pytest
* update
* updates
* fixes
* precommit fixes
* working multi-group SP
* fixing sample packing
* remove debug logs and simplify
* eval dataloader and sampler changes
* removing some obvious comments
* update config.qmd and rename option
* scoping down problematic import
* another import scoping change
* pernicious Fire CLI bugfix
* isolate cli tests
* actually isolate CLI tests
* gracefully handle no ring-flash-attn
* fix
* fix
* move ring flash attn to extras with flash-attn (#2414 )
* removing flash-attn from requirements.txt (in setup.py extras already)
* rename file, delete another
* using field validator instead of model validator
* test fix
* sampler / dataloader refactor
* non-seq2se1 collator fix
* removing print statement
* bugfix
* add SP doc, review comments
* small changes
* review comments, docstrings
* refactors, SP mixin
* small updates
* fix tests
* precommit
* precommit
---------
Co-authored-by: Wing Lian <wing.lian@gmail.com >
Co-authored-by: Dan Saunders <dan@axolotl.ai >
2025-03-21 12:43:55 -04:00
Dan Saunders
113e9cd193
Autodoc generation with quartodoc ( #2419 )
...
* quartodoc integration
* quartodoc progress
* deletions
* Update docs/.gitignore to exclude auto-generated API documentation files
* Fix
* more autodoc progress
* moving reference up near the top of the sidebar
* fix broken link
* update to reflect recent changes
* pydantic models refactor + add to autodoc + fixes
* fix
* shrinking header sizes
* fix accidental change
* include quartodoc build step
* update pre-commit version
* update pylint
* pre-commit
---------
Co-authored-by: Dan Saunders <dan@axolotl.ai >
2025-03-21 12:26:47 -04:00
NanoCode012
61825a464a
chore(doc): add explanation on fsdp_transformer_layer_cls_to_wrap ( #2429 ) [skip ci]
2025-03-21 11:59:22 -04:00
Dan Saunders
c907ac173e
adding pre-commit auto-update GH action and bumping plugin versions ( #2428 )
...
* adding pre-commit auto-update GH action and bumping plugin versions
* running updated pre-commit plugins
* sorry to revert, but pylint complained
* Update .pre-commit-config.yaml
Co-authored-by: Wing Lian <wing.lian@gmail.com >
---------
Co-authored-by: Dan Saunders <dan@axolotl.ai >
Co-authored-by: Wing Lian <wing.lian@gmail.com >
2025-03-21 11:02:43 -04:00
salman
187227d837
Fixing KTO+QLoRA+multi-GPU ( #2420 )
...
* WIP
* removing artifacts
* adding error
* adding adapter check
* linting
* simplifying check
* linting v2
* config fix -___-
2025-03-21 10:18:28 -04:00
NanoCode012
f8de8bb4f2
chore(doc): add instructions on adding custom integrations ( #2422 ) [skip ci]
...
* chore(doc): add instructions on adding custom integrations
* chore: add warning help
* feat: add note about integration path
* fix: adjust text per suggestion
2025-03-21 10:18:01 -04:00
hugo
8e604848a4
add run on novita ai ( #2421 ) [skip ci]
...
* add run on novita ai
* Revert "add run on novita ai"
This reverts commit 4d5df1ac6b
2025-03-21 10:17:47 -04:00
Wing Lian
aae4337f40
add 12.8.1 cuda to the base matrix ( #2426 )
...
* add 12.8.1 cuda to the base matrix
* use nightly
* bump deepspeed and set no binary
* deepspeed binary fixes hopefully
* install deepspeed by itself
* multiline fix
* make sure ninja is installed
* try with reversion of packaging/setuptools/wheel install
* use license instead of license-file
* try rolling back packaging and setuptools versions
* comment out license for validation for now
* make sure packaging version is consistent
* more parity across tests and docker images for packaging/setuptools
2025-03-21 10:17:25 -04:00
Wing Lian
38df5a36ea
bump HF versions except for trl ( #2427 )
2025-03-20 10:22:05 -04:00
Wing Lian
4d92a68a96
use default torch fused adamw optimizer as default as adamw_hf is deprecated ( #2425 )
...
* use default torch fused adamw optimizer as default as adamw_hf is deprecated
* make sure to have latest packaging installed
* bump packagingin requirements.txt too
2025-03-19 23:58:33 -04:00
SicariusSicariiStuff
85147ec430
Update README.md ( #2360 )
...
* Update README.md
wheel is needed
* feat: add ninja, setuptools, packing to installation steps
* fix: add missing instruction
---------
Co-authored-by: NanoCode012 <nano@axolotl.ai >
2025-03-17 08:39:17 -04:00
NanoCode012
51cd409488
Feat: minor docs improvements for RLHF and faq on embeddings ( #2401 ) [skip ci]
...
* feat: add doc on shrink_embeddings and custom calling
* chore: rename inference doc
* fix: clarify same config is used for all cli
* chore: rearrange order inference qmd
* feat: add simpo to doc
* fix: update defaults
* feat: add rl configs to doc
* fix: ensure beta consistent with trl.beta
* fix: clarify about lora/fft
* chore: rename title
* chore: fix language
* feat: move config reference higher
* Update docs/getting-started.qmd
Co-authored-by: salman <salman.mohammadi@outlook.com >
* Update docs/rlhf.qmd
Co-authored-by: salman <salman.mohammadi@outlook.com >
---------
Co-authored-by: salman <salman.mohammadi@outlook.com >
2025-03-17 08:39:04 -04:00
NanoCode012
7235123d44
chore(docs): add cookbook/blog link to docs ( #2410 ) [skip ci]
2025-03-17 08:38:19 -04:00
Wing Lian
4f5eb42a73
remove reference to deprecated import ( #2407 )
2025-03-15 08:49:41 -04:00
Wing Lian
fbe54be6b8
only validate hf user token on rank 0 ( #2408 )
2025-03-13 23:29:06 -04:00
Wing Lian
04f6324833
build cloud images with torch 2.6.0 ( #2413 )
...
* build cloud images with torch 2.6.0
* nightlies too
2025-03-13 23:28:51 -04:00
Wing Lian
f0072f3b9d
use max of 32 dataset processes if not explicit ( #2403 )
...
* use max of 32 dataset processes if not explicit
* change alternate min val for consistency
2025-03-11 12:02:58 -04:00
Wing Lian
59899b9817
pass additional info for fix untrained tokens when using distributed + offloading ( #2388 )
...
* pass additional info for fix untrained tokens when using distributed + offloading
* use latest version of vendored lib
* use v0.0.5 of contribs lgpl
* fix for no bad tokens and add tests
* use release
* add multigpu test too
* make sure the multigpu zero3 test actually uses zero3
2025-03-11 12:02:43 -04:00
NanoCode012
4a736986fa
fix(modal): add git pull when getting branch files ( #2399 )
2025-03-10 15:14:41 -04:00
Wing Lian
5d0f110a3b
include iproute2 and nvtop in cloud image ( #2393 )
2025-03-10 15:13:38 -04:00
NanoCode012
83f8698b8a
fix: create mount folder on modal if not exist ( #2390 )
2025-03-10 16:27:42 +07:00
xzuyn
60a11a6410
Use Latest Cut Cross Entropy ( #2392 )
...
* Update __init__.py
* Update README.md
* Update cutcrossentropy_install.py
* add test
2025-03-10 16:26:40 +07:00
NanoCode012
46a045e528
chore(doc): add faq when having no default chat_template ( #2398 )
...
* chore(doc): add faq when having no default chat_template
* Update docs/dataset-formats/conversation.qmd
Co-authored-by: salman <salman.mohammadi@outlook.com >
* Update docs/faq.qmd
Co-authored-by: salman <salman.mohammadi@outlook.com >
---------
Co-authored-by: salman <salman.mohammadi@outlook.com >
2025-03-10 16:25:50 +07:00
NanoCode012
3b477e08a0
feat(doc): add more info on RewardModel datasets ( #2391 )
...
* fix: reduce title size
* feat(doc): add rm dataset info
* Update docs/reward_modelling.qmd following suggestion
Co-authored-by: salman <salman.mohammadi@outlook.com >
---------
Co-authored-by: salman <salman.mohammadi@outlook.com >
2025-03-10 16:25:31 +07:00
NanoCode012
16dc6ee68d
refactor: trl grpo configs to have descriptions ( #2386 )
...
* refactor: trl grpo configs to have descriptions
* chore: caps
2025-03-07 08:58:53 -05:00
Wing Lian
fa7c79b3b9
remove lion-pytorch as it's already handled upstream ( #2389 )
2025-03-07 08:58:15 -05:00
Wing Lian
ae66374156
Optimizer refactor and add Muon support ( #2367 )
...
* add muon optimizer
optimizer_cls_and_kwargs is on trainer_kwargs
only add adamw_kwargs if they're non-null
fix mocks
better handling of override and check the optimizer
unwrap optimizer
* fix import
2025-03-06 11:49:19 -05:00
Wing Lian
5e21b1a9da
various fixes 20250305 ( #2384 )
...
* various validation fixes
* fix check for non-truthy value
2025-03-06 11:48:44 -05:00
mhenrichsen
575e5f28ec
Update Tokenizer Overrides Handling in models.py ( #1549 )
...
* override special tokens mock code
* fix(doc): remove duplicate config
* feat: replace added_tokens in tokenizer and add test
* make sure to run tokenizer modification on rank 0 only
* use is local main process instead
* feat: rename config
---------
Co-authored-by: NanoCode012 <nano@axolotl.ai >
Co-authored-by: Wing Lian <wing@axolotl.ai >
2025-03-05 11:15:12 -05:00
xzuyn
0134093acc
Add REX LR Scheduler ( #2380 )
...
* Update trainer_builder.py
* Update base.py
* Update __init__.py
* Update base.py
* Update base.py
* Update config.qmd
* Update base.py
* Update base.py
* Update base.py
* Update base.py
* Update base.py
* Update base.py
* Update base.py
* lint
* lint
* lint
* lint
* lint
* lint
* Update base.py
* Update base.py
* lint
* Update base.py
* Update base.py
* Move RexLR to `schedulers.py`
* Remove RexLR from `base.py`
* Fix tooltip formatting
* lint
* Create test_schedulers.py
* Use a default optimizer in test
* lint
* lint
* Add `warmup_steps` and `cosine_min_lr_ratio` to test
* lint
2025-03-05 10:26:11 -05:00
NanoCode012
d4de93a7bb
feat(grpo): add reward_weights config and refactor ( #2365 )
2025-03-05 10:02:08 -05:00
NanoCode012
c8191394e9
fix(doc): add missing low_cpu_mem_usage config to docs ( #2369 ) [skip ci]
2025-03-05 10:01:44 -05:00
NanoCode012
f18231c653
chore(doc): add clarification about mpi4py error on single gpu deepspeed ( #2383 ) [skip ci]
...
* chore(doc): add clarification about mpi4py error on single gpu deepspeed
* fix: lint
2025-03-05 10:01:28 -05:00
