Skip to content

[None][chore] add spec_decoding configs in perf benchmark scripts and fix typos #9533

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lancelly merged 1 commit into from
Nov 28, 2025
Merged

[None][chore] add spec_decoding configs in perf benchmark scripts and fix typos #9533

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
45 changes: 41 additions & 4 deletions tests/scripts/perf-sanity/run_benchmark_serve.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -215,6 +215,14 @@ def str_to_bool(value: str) -> bool:
"enable_padding": (True, str_to_bool),
}

# Nested under 'speculative_config' in YAML
SPECULATIVE_CONFIG_METRICS = {
"decoding_type": (True, str),
"max_draft_len": (True, int),
"speculative_model_dir": (True, str),
"eagle3_one_model": (True, str_to_bool),
}

CLIENT_CONFIG_METRICS = {
"concurrency": (False, int),
"iterations": (False, int),
Expand Down Expand Up @@ -250,6 +258,10 @@ def __init__(
enable_block_reuse: bool = False,
free_gpu_memory_fraction: float = 0.8,
enable_padding: bool = True,
decoding_type: str = "",
max_draft_len: int = 0,
speculative_model_dir: str = "",
eagle3_one_model: bool = False,
):
self.name = name
self.model_name = model_name
Expand All @@ -272,13 +284,16 @@ def __init__(
self.free_gpu_memory_fraction = free_gpu_memory_fraction
self.max_batch_size = max_batch_size
self.enable_padding = enable_padding
self.decoding_type = decoding_type
self.max_draft_len = max_draft_len
self.speculative_model_dir = speculative_model_dir
self.eagle3_one_model = eagle3_one_model

self.model_path = ""

def to_cmd(self, working_dir: str) -> List[str]:
model_dir = get_model_dir(self.model_name)
self.model_path = model_dir if os.path.exists(
model_dir) else self.model_name

def to_cmd(self, working_dir: str) -> List[str]:
config_path = os.path.join(working_dir,
f"extra-llm-api-config.{self.name}.yml")
return [
Expand All @@ -294,6 +309,7 @@ def generate_extra_llm_api_config(self) -> str:
f"moe_expert_parallel_size: {self.ep}",
f"pipeline_parallel_size: {self.pp}",
f"max_num_tokens: {self.max_num_tokens}",
f"max_batch_size: {self.max_batch_size}",
f"enable_attention_dp: {str(self.enable_attention_dp).lower()}",
f"disable_overlap_scheduler: {str(self.disable_overlap_scheduler).lower()}",
f"stream_interval: {self.stream_interval}",
Expand Down Expand Up @@ -324,6 +340,19 @@ def generate_extra_llm_api_config(self) -> str:
f" batching_wait_iters: {self.batching_wait_iters}")
config_lines.append(f" timeout_iters: {self.timeout_iters}")

# Add speculative_config if decoding_type is specified
if self.decoding_type:
config_lines.append("speculative_config:")
config_lines.append(f" decoding_type: {self.decoding_type}")
if self.max_draft_len > 0:
config_lines.append(f" max_draft_len: {self.max_draft_len}")
if self.speculative_model_dir:
config_lines.append(
f" speculative_model_dir: {self.speculative_model_dir}")
if self.eagle3_one_model:
config_lines.append(
f" eagle3_one_model: {str(self.eagle3_one_model).lower()}")

return "\n".join(config_lines)


Expand Down Expand Up @@ -467,7 +496,15 @@ def parse_config_file(config_file_path: str, select_pattern: str = None):
free_gpu_memory_fraction=server_config_data.get(
'free_gpu_memory_fraction', 0.8),
max_batch_size=server_config_data.get('max_batch_size', 256),
enable_padding=server_config_data.get('enable_padding', True))
enable_padding=server_config_data.get('enable_padding', True),
decoding_type=server_config_data.get('speculative_config',
{}).get('decoding_type', ''),
max_draft_len=server_config_data.get('speculative_config',
{}).get('max_draft_len', 0),
speculative_model_dir=server_config_data.get(
'speculative_config', {}).get('speculative_model_dir', ''),
eagle3_one_model=server_config_data.get(
'speculative_config', {}).get('eagle3_one_model', False))

server_id = len(server_configs)
server_configs.append(server_config)
Expand Down