[Misc] Remove deprecated code (#12383)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
@@ -100,7 +100,6 @@ class EngineArgs:
|
||||
kv_cache_dtype: str = 'auto'
|
||||
seed: int = 0
|
||||
max_model_len: Optional[int] = None
|
||||
worker_use_ray: bool = False
|
||||
# Note: Specifying a custom executor backend by passing a class
|
||||
# is intended for expert use only. The API may change without
|
||||
# notice.
|
||||
@@ -389,10 +388,6 @@ class EngineArgs:
|
||||
'to "ray" if Ray is installed and fail otherwise. Note that tpu '
|
||||
'only supports Ray for distributed inference.')
|
||||
|
||||
parser.add_argument(
|
||||
'--worker-use-ray',
|
||||
action='store_true',
|
||||
help='Deprecated, use ``--distributed-executor-backend=ray``.')
|
||||
parser.add_argument('--pipeline-parallel-size',
|
||||
'-pp',
|
||||
type=int,
|
||||
@@ -1071,7 +1066,6 @@ class EngineArgs:
|
||||
parallel_config = ParallelConfig(
|
||||
pipeline_parallel_size=self.pipeline_parallel_size,
|
||||
tensor_parallel_size=self.tensor_parallel_size,
|
||||
worker_use_ray=self.worker_use_ray,
|
||||
max_parallel_loading_workers=self.max_parallel_loading_workers,
|
||||
disable_custom_all_reduce=self.disable_custom_all_reduce,
|
||||
tokenizer_pool_config=TokenizerPoolConfig.create_config(
|
||||
|
||||
@@ -259,21 +259,6 @@ class Metrics:
|
||||
documentation="Number of emitted tokens.",
|
||||
labelnames=labelnames))
|
||||
|
||||
# Deprecated in favor of vllm:prompt_tokens_total
|
||||
self.gauge_avg_prompt_throughput = self._gauge_cls(
|
||||
name="vllm:avg_prompt_throughput_toks_per_s",
|
||||
documentation="Average prefill throughput in tokens/s.",
|
||||
labelnames=labelnames,
|
||||
multiprocess_mode="sum",
|
||||
)
|
||||
# Deprecated in favor of vllm:generation_tokens_total
|
||||
self.gauge_avg_generation_throughput = self._gauge_cls(
|
||||
name="vllm:avg_generation_throughput_toks_per_s",
|
||||
documentation="Average generation throughput in tokens/s.",
|
||||
labelnames=labelnames,
|
||||
multiprocess_mode="sum",
|
||||
)
|
||||
|
||||
|
||||
# end-metrics-definitions
|
||||
|
||||
@@ -635,20 +620,6 @@ class PrometheusStatLogger(StatLoggerBase):
|
||||
self._log_histogram(self.metrics.histogram_max_tokens_request,
|
||||
stats.max_tokens_requests)
|
||||
|
||||
def _log_prometheus_interval(self, prompt_throughput: float,
|
||||
generation_throughput: float) -> None:
|
||||
# Logs metrics to prometheus that are computed every logging_interval.
|
||||
# Support legacy gauge metrics that make throughput calculations on
|
||||
# the vLLM side. Moving forward, we should use counters like
|
||||
# counter_prompt_tokens, counter_generation_tokens
|
||||
# Which log raw data and calculate summaries using rate() on the
|
||||
# grafana/prometheus side. See
|
||||
# https://github.com/vllm-project/vllm/pull/2316#discussion_r1464204666
|
||||
self.metrics.gauge_avg_prompt_throughput.labels(
|
||||
**self.labels).set(prompt_throughput)
|
||||
self.metrics.gauge_avg_generation_throughput.labels(
|
||||
**self.labels).set(generation_throughput)
|
||||
|
||||
def log(self, stats: Stats):
|
||||
"""Logs to prometheus and tracked stats every iteration."""
|
||||
# Log to prometheus.
|
||||
@@ -664,20 +635,6 @@ class PrometheusStatLogger(StatLoggerBase):
|
||||
# Log locally every local_interval seconds.
|
||||
if local_interval_elapsed(stats.now, self.last_local_log,
|
||||
self.local_interval):
|
||||
# Compute summary metrics for tracked stats (and log them
|
||||
# to promethus if applicable).
|
||||
prompt_throughput = get_throughput(self.num_prompt_tokens,
|
||||
now=stats.now,
|
||||
last_log=self.last_local_log)
|
||||
generation_throughput = get_throughput(
|
||||
self.num_generation_tokens,
|
||||
now=stats.now,
|
||||
last_log=self.last_local_log)
|
||||
|
||||
self._log_prometheus_interval(
|
||||
prompt_throughput=prompt_throughput,
|
||||
generation_throughput=generation_throughput)
|
||||
|
||||
if self.spec_decode_metrics is not None:
|
||||
self._log_gauge(
|
||||
self.metrics.gauge_spec_decode_draft_acceptance_rate,
|
||||
|
||||
Reference in New Issue
Block a user