MCPcopy
hub / github.com/InternLM/lmdeploy / parse_args

Function parse_args

benchmark/profile_throughput.py:243–350  ·  view source on GitHub ↗
()

Source from the content-addressed store, hash-verified

241
242
243def parse_args():
244 parser = argparse.ArgumentParser(description='Benchmark the request throughput of lmdeploy '
245 'in localhost',
246 formatter_class=DefaultsAndTypesHelpFormatter)
247 parser.add_argument('dataset', type=str, help='the path dataset')
248 parser.add_argument('model_path',
249 type=str,
250 help='the path of the model in localhost or '
251 'the repo_id of the model in huggingface.co')
252 parser.add_argument('-c',
253 '--concurrency',
254 type=int,
255 help='Number of working threads to process the sampled prompts',
256 default=256)
257 parser.add_argument('-n', '--num-prompts', type=int, help='Number of prompts to process', default=5000)
258 parser.add_argument('--no-stream-output', action='store_true', help='Use stream output')
259 parser.add_argument('--skip-tokenize', action='store_true', help='Pre-tokenize input prompts before starting')
260 parser.add_argument('--skip-detokenize', action='store_true', help='Skip detokenizing output tokens')
261 parser.add_argument('--cancel-rate', type=float, help='Possibility of a request being canceled', default=0)
262 parser.add_argument('--use-uvloop', action='store_true')
263 parser.add_argument('--csv', type=str, help='Where to save the result.', default='./profile_throughput.csv')
264 parser.add_argument('--seed', type=int, default=0, help='Seed used in sampling prompts from dataset')
265 parser.add_argument('--distributed-executor-backend',
266 type=str,
267 default=None,
268 choices=['uni', 'mp', 'ray'],
269 help='backend of executor backend')
270 parser.add_argument('--dataset-name',
271 type=str,
272 default='sharegpt',
273 choices=['sharegpt', 'random'],
274 help='Name of the dataset to benchmark on.')
275 parser.add_argument(
276 '--sharegpt-output-len',
277 type=int,
278 default=None,
279 help='Output length for each request. Overrides the output length '
280 'from the ShareGPT dataset.',
281 )
282 parser.add_argument(
283 '--random-input-len',
284 type=int,
285 help='Number of input tokens per request, used only for random '
286 'dataset.',
287 )
288 parser.add_argument(
289 '--random-output-len',
290 type=int,
291 help='Number of output tokens per request, used only for random '
292 'dataset.',
293 )
294 parser.add_argument(
295 '--random-range-ratio',
296 type=float,
297 default=0.0,
298 help='Range of sampled ratio of input/output length, '
299 'used only for random dataset.',
300 )

Callers 1

mainFunction · 0.70

Calls 15

top_pMethod · 0.80
temperatureMethod · 0.80
backendMethod · 0.80
eager_modeMethod · 0.80
dllm_denoising_stepsMethod · 0.80
add_spec_groupMethod · 0.80
cache_max_entry_countMethod · 0.80
cache_block_seq_lenMethod · 0.80
enable_prefix_cachingMethod · 0.80
dpMethod · 0.80

Tested by

no test coverage detected