fix token state json and mistral tokenizer issue (#3522) [skip ci]

* fix token state json and mistral tokenizer issue

* centralize constants

* forgot to commit constants file

* Fix weakref in pickling relora state dict

* make curl a bit quieter so it doesn't log 2K lines

* fix path traversal for olmoe test

* more test fixes that weren't flagged previously

* chore: lint

* skip tests that fail b/c of OutOfResources

* scattermoe as slow tests

* update fbgemm-genai for torch 2.10
This commit is contained in:
Wing Lian
2026-03-21 22:46:10 -04:00
committed by GitHub
parent 2c05847a5f
commit 0ee98a0309
22 changed files with 249 additions and 57 deletions

View File

@@ -262,6 +262,7 @@ class TestDPOLlamaLora(unittest.TestCase):
train(cfg=cfg, dataset_meta=dataset_meta)
check_model_output_exists(Path(temp_dir) / "checkpoint-20", cfg)
@pytest.mark.skip(reason="TRL ORPO trainer has internal zip() length mismatch bug")
@with_temp_dir
def test_orpo_lora(self, temp_dir):
cfg = DictDefault(