A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 20:45 UTC
[Misc] Fix typos in spec. decode metrics logging. (#6470)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Thomas Parnell committed
Jul 16, 2024 at 15:55 UTC
7508a3dc34c2b7a5c5c971b13f15208d7dade442
1 file changed
+3
-3
vllm/engine/metrics.py
+3
-3
@@ -325,8 +325,8 @@ class LoggingStatLogger(StatLoggerBase):
325
f"System efficiency: {metrics.system_efficiency:.3f}, "
326
f"Number of speculative tokens: {metrics.num_spec_tokens}, "
327
f"Number of accepted tokens: {metrics.accepted_tokens}, "
328
- f"Number of draft tokens tokens: {metrics.draft_tokens}, "
329
- f"Number of emitted tokens tokens: {metrics.emitted_tokens}.")
328
+ f"Number of draft tokens: {metrics.draft_tokens}, "
329
+ f"Number of emitted tokens: {metrics.emitted_tokens}.")
330
331
332
class PrometheusStatLogger(StatLoggerBase):
@@ -457,4 +457,4 @@ class PrometheusStatLogger(StatLoggerBase):
457
458
class RayPrometheusStatLogger(PrometheusStatLogger):
459
"""RayPrometheusStatLogger uses Ray metrics instead."""
460
- _metrics_cls = RayMetrics
\ No newline at end of file
460
+ _metrics_cls = RayMetrics