MCPcopy
hub / github.com/z-lab/dflash / _make_decode_metrics

Function _make_decode_metrics

dflash/benchmark.py:112–117  ·  view source on GitHub ↗
(num_output_tokens: int, generation_tps: float, acceptance_lengths: list[int])

Source from the content-addressed store, hash-verified

110
111
112def _make_decode_metrics(num_output_tokens: int, generation_tps: float, acceptance_lengths: list[int]) -> SimpleNamespace:
113 return SimpleNamespace(
114 num_output_tokens=num_output_tokens,
115 time_per_output_token=1.0 / generation_tps if generation_tps > 0 else float("inf"),
116 acceptance_lengths=acceptance_lengths,
117 )
118
119
120def _print_decode_summary(responses: list[dict[int, SimpleNamespace]], block_size: int) -> None:

Callers 1

_run_mlxFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected