xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 00:06:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc][Tools][Benchmark] Add benchmark_serving supports for llama.cpp. (#18692)

Browse Source 
Signed-off-by: Duyi-Wang <duyi.wang@intel.com>
This commit is contained in:
Duyi-Wang 2025-05-29 20:02:08 +08:00 committed by GitHub
parent f8977c233f
commit b169d5f7b6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
benchmarks/backend_request_func.py
View File

@ -324,7 +324,7 @@ async def async_request_openai_completions(
most_recent_timestamp = timestamp most_recent_timestamp = timestamp
generated_text += text or "" generated_text += text or ""
elif usage := data.get("usage"): if usage := data.get("usage"):
output.output_tokens = usage.get("completion_tokens") output.output_tokens = usage.get("completion_tokens")
if first_chunk_received: if first_chunk_received:
output.success = True output.success = True
@ -611,6 +611,7 @@ ASYNC_REQUEST_FUNCS = {
"tensorrt-llm": async_request_trt_llm, "tensorrt-llm": async_request_trt_llm,
"scalellm": async_request_openai_completions, "scalellm": async_request_openai_completions,
"sglang": async_request_openai_completions, "sglang": async_request_openai_completions,
"llama.cpp": async_request_openai_completions,
} }
OPENAI_COMPATIBLE_BACKENDS = [ OPENAI_COMPATIBLE_BACKENDS = [

4
benchmarks/benchmark_serving.py
View File

@ -762,6 +762,10 @@ def main(args: argparse.Namespace):
if "temperature" not in sampling_params: if "temperature" not in sampling_params:
sampling_params["temperature"] = 0.0 # Default to greedy decoding. sampling_params["temperature"] = 0.0 # Default to greedy decoding.
if args.backend == "llama.cpp":
# Disable prompt caching in llama.cpp backend
sampling_params["cache_prompt"] = False
# Avoid GC processing "static" data - reduce pause times. # Avoid GC processing "static" data - reduce pause times.
gc.collect() gc.collect()
gc.freeze() gc.freeze()