tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

upgrade trl and accelerate (#3161)

Browse Source 
* upgrade trl==0.23.0

* upgrade accelerate patch fix

* add hints when using gradient_checkpointing with DPO

* set gradient-checpointing properly
This commit is contained in:
Wing Lian
2025-09-16 14:53:01 -04:00
committed by GitHub
parent d4cff1b7bb
commit 86d6ee7c05
4 changed files with 20 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/e2e/multigpu/test_llama.py
View File

@@ -199,7 +199,7 @@ class TestMultiGPULlama:
"max_steps": 2,
"micro_batch_size": 2,
"gradient_accumulation_steps": 2,
# "gradient_checkpointing": True,
"gradient_checkpointing": False,
"output_dir": temp_dir,
"dataset_prepared_path": temp_dir + "/last_run_prepared",
"warmup_steps": 0,
@@ -278,7 +278,7 @@ class TestMultiGPULlama:
"max_steps": 2,
"micro_batch_size": 2,
"gradient_accumulation_steps": 2,
# "gradient_checkpointing": True,
"gradient_checkpointing": False,
"output_dir": temp_dir,
"dataset_prepared_path": temp_dir + "/last_run_prepared",
"warmup_steps": 0,