pull/570/head
Artem Chumachenko 2 months ago
parent 63a421bad9
commit 5ca2a031f5

@ -142,8 +142,8 @@ def test_sampling(tokenizer, model, ref_model, max_new_tokens=10):
@pytest.mark.skipif(
MODEL_NAME.lower().find("bloom") == -1,
reason="Mixtral and Llama uses DynamicCache, which can change based on beam search choices",
"bloom" not in MODEL_NAME.lower(),
reason="Mixtral and Llama use DynamicCache, which can change based on beam search choices",
)
@pytest.mark.forked
def test_beam_search_generation(tokenizer, model, ref_model, max_new_tokens=4, num_beams=5):

Loading…
Cancel
Save