Wing Lian 1cf7075d18 support seperate lr for embeddings, similar to loraplus (#1910) [skip ci] ...

* support seperate lr for embeddings, similar to loraplus

* add test case for train w lr embedding scale

* use kwarg for optimizer

* make sure to handle the optimizer creation

* make sure to handle for embedding_lr too

* use smollm for e2e, check for embeddings lr first before wdecay

2024-11-29 20:38:20 -05:00

..

integrations

upgrade liger to 0.4.0 (#1973)

2024-11-07 12:53:34 -05:00

multigpu

add finetome dataset to fixtures, check eval_loss in test (#2106) [skip ci]

2024-11-29 20:37:32 -05:00

patched

bump transformers for fsdp-grad-accum fix, remove patch (#2079)

2024-11-19 02:23:09 -05:00

__init__.py

missing dunder-init

2023-11-06 18:33:01 -05:00

.gitignore

Support Sample packing for phi arch (#586)

2023-09-15 15:46:54 -04:00

test_dpo.py

Add weighted optimisation support for trl DPO trainer integration (#2016)

2024-11-08 11:29:11 -05:00

test_embeddings_lr.py

support seperate lr for embeddings, similar to loraplus (#1910) [skip ci]

2024-11-29 20:38:20 -05:00

test_falcon.py

Falcon embeddings (#1149) [skip docker]

2024-01-22 21:01:42 -05:00

test_imports.py

Bump deepspeed 20240727 (#1790)

2024-07-27 10:24:11 -04:00

test_llama_pretrain.py

fixes to accelerator so that iterable pretraining datasets work (#1759)

2024-07-17 10:58:38 -04:00

test_llama.py

Update get_unpad_data patching for multipack (#2013)

2024-11-15 20:35:50 -05:00

test_load_model.py

Refactor func load_model to class ModelLoader (#1909)

2024-10-25 09:06:56 -04:00

test_lora_llama.py

add basic support for the optimi adamw optimizer (#1727)

2024-07-14 19:12:57 -04:00

test_mamba.py

attempt to also run e2e tests that needs gpus (#1070)

2024-01-09 21:23:23 -05:00

test_mistral.py

use temp_dir kwarg instead

2023-11-06 18:33:01 -05:00

test_mixtral.py

fix broken linting (#1541)

2024-04-19 01:03:04 -04:00

test_optimizers.py

optim e2e tests to run a bit faster (#2069) [skip ci]

2024-11-18 12:35:31 -05:00

test_packing_loss.py

optim e2e tests to run a bit faster (#2069) [skip ci]

2024-11-18 12:35:31 -05:00

test_phi.py

re-enable phi for tests in modal ci (#1373)

2024-05-29 15:41:46 -04:00

test_qwen.py

remove the bos token from dpo outputs (#1733) [skip ci]

2024-11-15 19:09:20 -05:00

test_relora_llama.py

relora: magnitude pruning of the optimizer (#1245)

2024-02-06 00:35:30 -05:00

test_reward_model_llama.py

Reward model (#1879)

2024-10-13 15:11:13 -04:00

utils.py

ADOPT optimizer integration (#2032) [skip ci]

2024-11-13 17:10:17 -05:00