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 8, 01:15 UTC
fix(docs): fix typos in comments and docstrings (#34836)
Signed-off-by: machov <mv1742@nyu.edu>
Manrique Vargas committed
Feb 19, 2026 at 02:17 UTC
ad5aa6bd9f4394cf48d794ac09bfff911170b557
5 files changed
+6
-6
vllm/compilation/backends.py
+1
-1
@@ -285,7 +285,7 @@ class CompilerManager:
285
with self.compile_context(compile_range):
286
# There is a compilation time optimization here.
287
#
288
- # If the (input metdata, graph, compiler config) are the same, then
288
+ # If the (input metadata, graph, compiler config) are the same, then
289
# we want to avoid compiling the same artifact again. If we didn't
290
# do this optimization, the backend compilation (InductorAdaptor or
291
# InductorStandaloneAdaptor)
vllm/model_executor/layers/fused_moe/oracle/fp8.py
+1
-1
@@ -420,7 +420,7 @@ def make_fp8_moe_quant_config(
420
per_out_ch_quant: bool = False,
421
) -> FusedMoEQuantConfig | None:
422
"""
423
- Create FusedMoEQuantConfig for the specifed FP8 Backend.
423
+ Create FusedMoEQuantConfig for the specified FP8 Backend.
424
The FusedMoEQuantConfig holds the scales that are used
425
at runtime by the Modular Kernel abstraction.
426
vllm/model_executor/layers/fused_moe/runner/default_moe_runner.py
+1
-1
@@ -151,7 +151,7 @@ class DefaultMoERunner(MoERunner):
151
kernels for different parallel execution modes.
152
153
Eventually, this class will be split up and specialized for different
154
- configurations, e.g. the presense or absence of shared experts, a gate, etc.
154
+ configurations, e.g. the presence or absence of shared experts, a gate, etc.
155
"""
156
157
def __init__(
vllm/model_executor/models/gpt_oss.py
+1
-1
@@ -586,7 +586,7 @@ class GptOssModel(nn.Module):
586
parts = name.split(".")
587
ids = [s for s in parts if s.isdigit()]
588
589
- # for amd-quark format that each expert is seperated
589
+ # for amd-quark format that each expert is separated
590
# need to extract the parameter name with experts fused.
591
# example model: amd/gpt-oss-20b-MoE-Quant-W-MXFP4-A-FP8-KV-FP8
592
if len(ids) == 2:
vllm/utils/torch_utils.py
+2
-2
@@ -567,8 +567,8 @@ def current_stream() -> torch.cuda.Stream:
567
return _current_stream_tls.value
568
569
570
-# Global auxilary stream for running operations in background streams.
571
-# We have single global auxilary stream to avoid an explosion of streams
570
+# Global auxiliary stream for running operations in background streams.
571
+# We have single global auxiliary stream to avoid an explosion of streams
572
# for every layer (and make profiling look sane).
573
#
574
# aux_stream() is currently used for: