Skip to content

Commit

Permalink
Remove eos_token_id
Browse files Browse the repository at this point in the history
  • Loading branch information
l-bat committed Jan 7, 2025
1 parent 7eb1e54 commit 6c3e950
Showing 1 changed file with 2 additions and 4 deletions.
6 changes: 2 additions & 4 deletions tests/python_tests/test_kv_cache_eviction.py
Original file line number Diff line number Diff line change
Expand Up @@ -203,9 +203,9 @@ class LongBenchTestData:

@pytest.mark.precommit
@pytest.mark.parametrize("test_struct", [
LongBenchTestData("samsum", 36.78, 14, 9.596),
LongBenchTestData("samsum", 37.84, 11.8, 7.68),
LongBenchTestData("trec", 28.12, 11.8, 7.721),
LongBenchTestData("qasper", 21.68, 18.4, 12.706),
LongBenchTestData("qasper", 15.88, 11.8, 6.483),
])
def test_optimized_generation_longbench(qwen2_converted_model, test_struct):
seqs_per_request = 32
Expand All @@ -215,12 +215,10 @@ def test_optimized_generation_longbench(qwen2_converted_model, test_struct):
model_name = "/".join(models_path.parts[-2:])
subset = test_struct.subset
max_new_tokens = dataset2maxlen[subset]
tokenizer = qwen2_converted_model.tokenizer

generation_config = GenerationConfig() # expecting default greedy sampling
generation_config.num_return_sequences = 1
generation_config.max_new_tokens = max_new_tokens
generation_config.eos_token_id = tokenizer.eos_token_id

scheduler_config.use_cache_eviction = True
if scheduler_config.use_cache_eviction:
Expand Down

0 comments on commit 6c3e950

Please sign in to comment.