jagged lr restart scheudler (#1680) [skip ci]

* jagged lr restart scheudler

var name fix
make sure to create scheduler first

* wire things together

* more fixes

* fix for nesting scheduler and first anneal phase

* no need for relora trainer anymore since we've generalized the relora scheduler

* remove redundant relora scheduler and lint

* update relora e2e test for updated params

* need restart steps for relora test

* update quarto docs for dropped relora trainer

* update example yaml

* drop verbose arg

* min lr scale support for jagged lr

* don't let min_lr be nonetype

* cleanup args
This commit is contained in:
Wing Lian
2025-07-31 13:50:03 -04:00
committed by GitHub
parent 32a7890231
commit 7b68dfafd7
15 changed files with 139 additions and 137 deletions

View File

@@ -34,9 +34,10 @@ class TestReLoraLlama(unittest.TestCase):
"lora_alpha": 16,
"lora_dropout": 0.05,
"lora_target_modules": ["q_proj", "v_proj"],
"relora_steps": 50,
"relora_warmup_steps": 10,
"relora_anneal_steps": 10,
"relora": True,
"jagged_restart_steps": 50,
"jagged_restart_warmup_steps": 10,
"jagged_restart_anneal_steps": 10,
"relora_prune_ratio": 0.9,
"relora_cpu_offload": True,
"val_set_size": 0.0,