biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
tests/lora/test_chatglm3_tp.py
View File

@@ -21,20 +21,18 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> list[str]:
prompts = [
PROMPT_TEMPLATE.format(query="How many singers do we have?"),
PROMPT_TEMPLATE.format(
query=
"What is the average, minimum, and maximum age of all singers from France?" # noqa: E501
query="What is the average, minimum, and maximum age of all singers from France?" # noqa: E501
),
PROMPT_TEMPLATE.format(
query=
"Show name, country, age for all singers ordered by age from the oldest to the youngest." # noqa: E501
query="Show name, country, age for all singers ordered by age from the oldest to the youngest." # noqa: E501
),
]
sampling_params = vllm.SamplingParams(temperature=0, max_tokens=32)
outputs = llm.generate(
prompts,
sampling_params,
lora_request=LoRARequest(str(lora_id), lora_id, lora_path)
if lora_id else None)
lora_request=LoRARequest(str(lora_id), lora_id, lora_path) if lora_id else None,
)
# Print the outputs.
generated_texts: list[str] = []
for output in outputs:
@@ -47,13 +45,15 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> list[str]:
@create_new_process_for_each_test()
def test_chatglm3_lora(chatglm3_lora_files):
llm = vllm.LLM(MODEL_PATH,
max_model_len=1024,
enable_lora=True,
max_loras=4,
max_lora_rank=64,
trust_remote_code=True,
enable_chunked_prefill=True)
llm = vllm.LLM(
MODEL_PATH,
max_model_len=1024,
enable_lora=True,
max_loras=4,
max_lora_rank=64,
trust_remote_code=True,
enable_chunked_prefill=True,
)
output1 = do_sample(llm, chatglm3_lora_files, lora_id=1)
for i in range(len(EXPECTED_LORA_OUTPUT)):
@@ -66,15 +66,17 @@ def test_chatglm3_lora(chatglm3_lora_files):
@multi_gpu_test(num_gpus=4)
@create_new_process_for_each_test()
def test_chatglm3_lora_tp4(chatglm3_lora_files):
llm = vllm.LLM(MODEL_PATH,
max_model_len=1024,
enable_lora=True,
max_loras=4,
max_lora_rank=64,
tensor_parallel_size=4,
trust_remote_code=True,
fully_sharded_loras=False,
enable_chunked_prefill=True)
llm = vllm.LLM(
MODEL_PATH,
max_model_len=1024,
enable_lora=True,
max_loras=4,
max_lora_rank=64,
tensor_parallel_size=4,
trust_remote_code=True,
fully_sharded_loras=False,
enable_chunked_prefill=True,
)
output1 = do_sample(llm, chatglm3_lora_files, lora_id=1)
for i in range(len(EXPECTED_LORA_OUTPUT)):
@@ -90,16 +92,18 @@ def test_chatglm3_lora_tp4_fully_sharded_loras(chatglm3_lora_files):
# https://github.com/NVIDIA/nccl/issues/1790, set a lower value for
# gpu_memory_utilization here because NCCL >= 2.26.3 seems to use
# more GPU memory causing vLLM to OOM
llm = vllm.LLM(MODEL_PATH,
max_model_len=1024,
enable_lora=True,
max_loras=4,
max_lora_rank=64,
tensor_parallel_size=4,
trust_remote_code=True,
fully_sharded_loras=True,
enable_chunked_prefill=True,
gpu_memory_utilization=0.85)
llm = vllm.LLM(
MODEL_PATH,
max_model_len=1024,
enable_lora=True,
max_loras=4,
max_lora_rank=64,
tensor_parallel_size=4,
trust_remote_code=True,
fully_sharded_loras=True,
enable_chunked_prefill=True,
gpu_memory_utilization=0.85,
)
output1 = do_sample(llm, chatglm3_lora_files, lora_id=1)
for i in range(len(EXPECTED_LORA_OUTPUT)):
assert output1[i] == EXPECTED_LORA_OUTPUT[i]