@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 43c146ca42

Fork 0

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 14, 05:30 UTC

[Misc] Clean up unnecessary E501 ignore (#26274)

Signed-off-by: Roger Wang <hey@rogerw.io>

Roger Wang committed Oct 6, 2025 at 00:29 UTC 43c146ca4286efb49f1f15dc46b136b69ee7e684

8 files changed +38 -38

vllm/benchmarks/datasets.py

+2 -2

@@ -626,7 +626,7 @@ class RandomDataset(BenchmarkDataset):

626 # Decode, then re-encode and truncate to preserve token count invariants

627 total_input_len = prefix_len + int(input_len)

628 prompt, adjusted_token_sequence, token_mismatch = (

629 - gen_prompt_decode_to_target_len( # noqa: E501

629 + gen_prompt_decode_to_target_len(

630 tokenizer=tokenizer,

631 token_sequence=token_sequence,

632 target_token_len=total_input_len,

@@ -2855,7 +2855,7 @@ class PrefixRepetitionRandomDataset(BenchmarkDataset):

2855 for _ in range(prompts_per_prefix):

2856 suffix_tokens, token_mistmatch = _generate_exact_length_tokens(

2857 suffix_len

2858 - ) # noqa: E501

2858 + )

2859 token_mismatch_total += token_mistmatch

2860 combined_tokens = prefix_tokens + suffix_tokens

2861 prompt = tokenizer.decode(combined_tokens)

vllm/benchmarks/throughput.py

+2 -2

@@ -459,14 +459,14 @@ def validate_args(args):

459 ):

460 assert args.backend == "vllm-chat", (

461 f"{args.dataset_path} needs to use vllm-chat as the backend."

462 - ) # noqa: E501

462 + )

463 elif args.dataset_path in (

464 InstructCoderDataset.SUPPORTED_DATASET_PATHS

465 | AIMODataset.SUPPORTED_DATASET_PATHS

466 ):

467 assert args.backend == "vllm", (

468 f"{args.dataset_path} needs to use vllm as the backend."

469 - ) # noqa: E501

469 + )

470 else:

471 raise ValueError(f"{args.dataset_path} is not supported by hf dataset.")

472

vllm/compilation/inductor_pass.py

+1 -1

@@ -19,7 +19,7 @@ if is_torch_equal_or_newer("2.6"):

19 from torch._inductor.custom_graph_pass import CustomGraphPass

20 else:

21 # CustomGraphPass is not present in 2.5 or lower, import our version

22 - from .torch25_custom_graph_pass import ( # noqa: E501

22 + from .torch25_custom_graph_pass import (

23 Torch25CustomGraphPass as CustomGraphPass,

24 )

vllm/model_executor/layers/quantization/__init__.py

+1 -1

@@ -95,7 +95,7 @@ def get_quantization_config(quantization: str) -> type[QuantizationConfig]:

95 from .awq_marlin import AWQMarlinConfig

96 from .bitblas import BitBLASConfig

97 from .bitsandbytes import BitsAndBytesConfig

98 - from .compressed_tensors.compressed_tensors import ( # noqa: E501

98 + from .compressed_tensors.compressed_tensors import (

99 CompressedTensorsConfig,

100 )

101 from .deepspeedfp import DeepSpeedFPConfig

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py

+2 -2

@@ -26,7 +26,7 @@ from vllm.model_executor.layers.linear import (

26 UnquantizedLinearMethod,

27 )

28 from vllm.model_executor.layers.quantization import QuantizationMethods

29 -from vllm.model_executor.layers.quantization.base_config import ( # noqa: E501

29 +from vllm.model_executor.layers.quantization.base_config import (

30 QuantizationConfig,

31 QuantizeMethodBase,

32 )

@@ -256,7 +256,7 @@ class CompressedTensorsConfig(QuantizationConfig):

256 )

257 else:

258 target_scheme_map[target]["input_activations"] = (

259 - QuantizationArgs.model_validate( # noqa: E501

259 + QuantizationArgs.model_validate(

260 quant_config.get("input_activations")

261 )

262 )

vllm/model_executor/models/gemma3n_mm.py

+1 -1

@@ -176,7 +176,7 @@ class Gemma3nDummyInputsBuilder(BaseDummyInputsBuilder[Gemma3nProcessingInfo]):

176 processor = self.info.get_hf_processor()

177 audio_feature_extractor: Gemma3nAudioFeatureExtractor = (

178 processor.feature_extractor

179 - ) # noqa: E501

179 + )

180 audio_len = audio_feature_extractor.fft_length

181 image_processor: SiglipImageProcessorFast = processor.image_processor

182 img_width = image_processor.size.get("width", 224)

vllm/model_executor/models/registry.py

+28 -28

@@ -120,7 +120,7 @@ _TEXT_GENERATION_MODELS = {

120 "JambaForCausalLM": ("jamba", "JambaForCausalLM"),

121 "Lfm2ForCausalLM": ("lfm2", "Lfm2ForCausalLM"),

122 "LlamaForCausalLM": ("llama", "LlamaForCausalLM"),

123 - "Llama4ForCausalLM": ("llama4", "Llama4ForCausalLM"), # noqa: E501

123 + "Llama4ForCausalLM": ("llama4", "Llama4ForCausalLM"),

124 # For decapoda-research/llama-*

125 "LLaMAForCausalLM": ("llama", "LlamaForCausalLM"),

126 "LongcatFlashForCausalLM": ("longcat_flash", "LongcatFlashForCausalLM"),

@@ -204,7 +204,7 @@ _EMBEDDING_MODELS = {

204 "LlavaNextForConditionalGeneration": (

205 "llava_next",

206 "LlavaNextForConditionalGeneration",

207 - ), # noqa: E501

207 + ),

208 "Phi3VForCausalLM": ("phi3v", "Phi3VForCausalLM"),

209 "Qwen2VLForConditionalGeneration": ("qwen2_vl", "Qwen2VLForConditionalGeneration"), # noqa: E501

210 # Technically Terratorch models work on images, both in

@@ -240,46 +240,46 @@ _MULTIMODAL_MODELS = {

240 "AyaVisionForConditionalGeneration": (

241 "aya_vision",

242 "AyaVisionForConditionalGeneration",

243 - ), # noqa: E501

243 + ),

244 "Blip2ForConditionalGeneration": ("blip2", "Blip2ForConditionalGeneration"),

245 "ChameleonForConditionalGeneration": (

246 "chameleon",

247 "ChameleonForConditionalGeneration",

248 - ), # noqa: E501

248 + ),

249 "Cohere2VisionForConditionalGeneration": (

250 "cohere2_vision",

251 "Cohere2VisionForConditionalGeneration",

252 - ), # noqa: E501

252 + ),

253 "DeepseekVLV2ForCausalLM": ("deepseek_vl2", "DeepseekVLV2ForCausalLM"),

254 "DotsOCRForCausalLM": ("dots_ocr", "DotsOCRForCausalLM"),

255 "Ernie4_5_VLMoeForConditionalGeneration": (

256 "ernie45_vl",

257 "Ernie4_5_VLMoeForConditionalGeneration",

258 - ), # noqa: E501

258 + ),

259 "FuyuForCausalLM": ("fuyu", "FuyuForCausalLM"),

260 "Gemma3ForConditionalGeneration": ("gemma3_mm", "Gemma3ForConditionalGeneration"), # noqa: E501

261 "Gemma3nForConditionalGeneration": (

262 "gemma3n_mm",

263 "Gemma3nForConditionalGeneration",

264 - ), # noqa: E501

264 + ),

265 "GLM4VForCausalLM": ("glm4v", "GLM4VForCausalLM"),

266 "Glm4vForConditionalGeneration": ("glm4_1v", "Glm4vForConditionalGeneration"), # noqa: E501

267 "Glm4vMoeForConditionalGeneration": ("glm4_1v", "Glm4vMoeForConditionalGeneration"), # noqa: E501

268 "GraniteSpeechForConditionalGeneration": (

269 "granite_speech",

270 "GraniteSpeechForConditionalGeneration",

271 - ), # noqa: E501

271 + ),

272 "H2OVLChatModel": ("h2ovl", "H2OVLChatModel"),

273 "InternVLChatModel": ("internvl", "InternVLChatModel"),

274 "NemotronH_Nano_VL_V2": ("nano_nemotron_vl", "NemotronH_Nano_VL_V2"),

275 "InternS1ForConditionalGeneration": (

276 "interns1",

277 "InternS1ForConditionalGeneration",

278 - ), # noqa: E501

278 + ),

279 "InternVLForConditionalGeneration": (

280 "interns1",

281 "InternS1ForConditionalGeneration",

282 - ), # noqa: E501

282 + ),

283 "Idefics3ForConditionalGeneration": (

284 "idefics3",

285 "Idefics3ForConditionalGeneration",

@@ -289,7 +289,7 @@ _MULTIMODAL_MODELS = {

289 "KeyeVL1_5ForConditionalGeneration": (

290 "keye_vl1_5",

291 "KeyeVL1_5ForConditionalGeneration",

292 - ), # noqa: E501

292 + ),

293 "RForConditionalGeneration": ("rvl", "RForConditionalGeneration"),

294 "KimiVLForConditionalGeneration": ("kimi_vl", "KimiVLForConditionalGeneration"), # noqa: E501

295 "Llama_Nemotron_Nano_VL": ("nemotron_vl", "LlamaNemotronVLChatModel"),

@@ -298,27 +298,27 @@ _MULTIMODAL_MODELS = {

298 "LlavaNextForConditionalGeneration": (

299 "llava_next",

300 "LlavaNextForConditionalGeneration",

301 - ), # noqa: E501

301 + ),

302 "LlavaNextVideoForConditionalGeneration": (

303 "llava_next_video",

304 "LlavaNextVideoForConditionalGeneration",

305 - ), # noqa: E501

305 + ),

306 "LlavaOnevisionForConditionalGeneration": (

307 "llava_onevision",

308 "LlavaOnevisionForConditionalGeneration",

309 - ), # noqa: E501

309 + ),

310 "MantisForConditionalGeneration": ("llava", "MantisForConditionalGeneration"), # noqa: E501

311 "MiDashengLMModel": ("midashenglm", "MiDashengLMModel"),

312 "MiniMaxVL01ForConditionalGeneration": (

313 "minimax_vl_01",

314 "MiniMaxVL01ForConditionalGeneration",

315 - ), # noqa: E501

315 + ),

316 "MiniCPMO": ("minicpmo", "MiniCPMO"),

317 "MiniCPMV": ("minicpmv", "MiniCPMV"),

318 "Mistral3ForConditionalGeneration": (

319 "mistral3",

320 "Mistral3ForConditionalGeneration",

321 - ), # noqa: E501

321 + ),

322 "MolmoForCausalLM": ("molmo", "MolmoForCausalLM"),

323 "NVLM_D": ("nvlm_d", "NVLM_D_Model"),

324 "Ovis": ("ovis", "Ovis"),

@@ -326,7 +326,7 @@ _MULTIMODAL_MODELS = {

326 "PaliGemmaForConditionalGeneration": (

327 "paligemma",

328 "PaliGemmaForConditionalGeneration",

329 - ), # noqa: E501

329 + ),

330 "Phi3VForCausalLM": ("phi3v", "Phi3VForCausalLM"),

331 "Phi4MMForCausalLM": ("phi4mm", "Phi4MMForCausalLM"),

332 "Phi4MultimodalForCausalLM": ("phi4_multimodal", "Phi4MultimodalForCausalLM"), # noqa: E501

@@ -336,31 +336,31 @@ _MULTIMODAL_MODELS = {

336 "Qwen2_5_VLForConditionalGeneration": (

337 "qwen2_5_vl",

338 "Qwen2_5_VLForConditionalGeneration",

339 - ), # noqa: E501

339 + ),

340 "Qwen2AudioForConditionalGeneration": (

341 "qwen2_audio",

342 "Qwen2AudioForConditionalGeneration",

343 - ), # noqa: E501

343 + ),

344 "Qwen2_5OmniModel": (

345 "qwen2_5_omni_thinker",

346 "Qwen2_5OmniThinkerForConditionalGeneration",

347 - ), # noqa: E501

347 + ),

348 "Qwen2_5OmniForConditionalGeneration": (

349 "qwen2_5_omni_thinker",

350 "Qwen2_5OmniThinkerForConditionalGeneration",

351 - ), # noqa: E501

351 + ),

352 "Qwen3VLForConditionalGeneration": ("qwen3_vl", "Qwen3VLForConditionalGeneration"), # noqa: E501

353 "Qwen3VLMoeForConditionalGeneration": (

354 "qwen3_vl_moe",

355 "Qwen3VLMoeForConditionalGeneration",

356 - ), # noqa: E501

356 + ),

357 "SkyworkR1VChatModel": ("skyworkr1v", "SkyworkR1VChatModel"),

358 "Step3VLForConditionalGeneration": ("step3_vl", "Step3VLForConditionalGeneration"), # noqa: E501

359 "TarsierForConditionalGeneration": ("tarsier", "TarsierForConditionalGeneration"), # noqa: E501

360 "Tarsier2ForConditionalGeneration": (

361 "qwen2_vl",

362 "Tarsier2ForConditionalGeneration",

363 - ), # noqa: E501

363 + ),

364 "UltravoxModel": ("ultravox", "UltravoxModel"),

365 "VoxtralForConditionalGeneration": ("voxtral", "VoxtralForConditionalGeneration"), # noqa: E501

366 # [Encoder-decoder]

@@ -401,23 +401,23 @@ _TRANSFORMERS_BACKEND_MODELS = {

401 "TransformersMoEForMultimodalLM": (

402 "transformers_moe",

403 "TransformersMoEForMultimodalLM",

404 - ), # noqa: E501

404 + ),

405 "TransformersEmbeddingModel": (

406 "transformers_pooling",

407 "TransformersEmbeddingModel",

408 - ), # noqa: E501

408 + ),

409 "TransformersForSequenceClassification": (

410 "transformers_pooling",

411 "TransformersForSequenceClassification",

412 - ), # noqa: E501

412 + ),

413 "TransformersMoEForSequenceClassification": (

414 "transformers_pooling",

415 "TransformersMoEForSequenceClassification",

416 - ), # noqa: E501

416 + ),

417 "TransformersMoEEmbeddingModel": (

418 "transformers_pooling",

419 "TransformersMoEEmbeddingModel",

420 - ), # noqa: E501

420 + ),

421 }

422

423 _VLLM_MODELS = {

vllm/v1/attention/backends/gdn_attn.py

+1 -1

@@ -79,7 +79,7 @@ class GDNAttentionMetadataBuilder(AttentionMetadataBuilder[GDNAttentionMetadata]

79 self.speculative_config = vllm_config.speculative_config

80 self.kv_cache_spec = kv_cache_spec

81 if self.speculative_config:

82 - self.num_spec = self.speculative_config.num_speculative_tokens # noqa: E501

82 + self.num_spec = self.speculative_config.num_speculative_tokens

83 else:

84 self.num_spec = 0

85 self.use_spec_decode = self.num_spec > 0