[Quantization / tests ] Fix bnb MPT test (#27178)

fix bnb mpt test
This commit is contained in:
Younes Belkada
2023-10-31 16:25:53 +01:00
committed by GitHub
parent 05f2290114
commit 4bb50aa212

View File

@@ -124,7 +124,6 @@ class MixedInt8Test(BaseMixedInt8Test):
gc.collect()
torch.cuda.empty_cache()
@unittest.skip("Un-skip once https://github.com/mosaicml/llm-foundry/issues/703 is resolved")
def test_get_keys_to_not_convert_trust_remote_code(self):
r"""
Test the `get_keys_to_not_convert` function with `trust_remote_code` models.
@@ -135,11 +134,11 @@ class MixedInt8Test(BaseMixedInt8Test):
model_id = "mosaicml/mpt-7b"
config = AutoConfig.from_pretrained(
model_id, trust_remote_code=True, revision="72e5f594ce36f9cabfa2a9fd8f58b491eb467ee7"
model_id, trust_remote_code=True, revision="ada218f9a93b5f1c6dce48a4cc9ff01fcba431e7"
)
with init_empty_weights():
model = AutoModelForCausalLM.from_config(
config, trust_remote_code=True, code_revision="72e5f594ce36f9cabfa2a9fd8f58b491eb467ee7"
config, trust_remote_code=True, code_revision="ada218f9a93b5f1c6dce48a4cc9ff01fcba431e7"
)
self.assertEqual(get_keys_to_not_convert(model), ["transformer.wte"])