Multipack simplify for Mixtral (#1142)

This commit is contained in:
Wing Lian
2024-01-18 16:23:49 -05:00
committed by GitHub
parent 1d70f24b50
commit 6910e6a8ca
11 changed files with 201 additions and 430 deletions

View File

@@ -52,11 +52,7 @@ class TestModelPatches(unittest.TestCase):
model, _ = load_model(cfg, tokenizer, inference=cli_args.inference)
assert (
"axolotl.monkeypatch.mixtral.modeling_mixtral"
in model.model.layers[0].self_attn.__class__.__module__
)
assert (
"MixtralMultipackFlashAttention2"
"MixtralFlashAttention2"
in model.model.layers[0].self_attn.__class__.__name__
)