@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / c38b8d5a31

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 7, 20:15 UTC

Remove `padding_index` from models that don't use it for better Transformers v5 compatibility (#35189)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

Harry Mellor committed Feb 24, 2026 at 16:04 UTC c38b8d5a317da356353a0a9ae1ab87f612267fb3

14 files changed -14

vllm/model_executor/models/ernie45_moe.py

-1

@@ -421,7 +421,6 @@ class Ernie4_5_MoeModel(nn.Module):

421 cache_config = vllm_config.cache_config

422 quant_config = vllm_config.quant_config

423

424 - self.padding_idx = config.pad_token_id

424 self.vocab_size = config.vocab_size

425 self.config = config

426 parallel_config = vllm_config.parallel_config

vllm/model_executor/models/ernie45_vl_moe.py

-1

@@ -523,7 +523,6 @@ class Ernie4_5_VLMoeModel(nn.Module):

523 cache_config = vllm_config.cache_config

524 quant_config = vllm_config.quant_config

525

526 - self.padding_idx = config.pad_token_id

526 self.vocab_size = config.vocab_size

527 self.config = config

528

vllm/model_executor/models/granitemoeshared.py

-1

@@ -157,7 +157,6 @@ class GraniteMoeSharedModel(nn.Module):

157

158 self.config = config

159 self.quant_config = quant_config # Required by MixtralModel

160 - self.padding_idx = config.pad_token_id

160

161 self.vocab_size = config.vocab_size

162

vllm/model_executor/models/grok1.py

-1

@@ -451,7 +451,6 @@ class Grok1Model(nn.Module):

451

452 self.config = config

453 self.quant_config = quant_config

454 - self.padding_idx = config.pad_token_id

454

455 # Store expert naming for weight loading

456 self.ckpt_gate_proj_name = ckpt_gate_proj_name

vllm/model_executor/models/hunyuan_v1.py

-1

@@ -600,7 +600,6 @@ class HunYuanModel(nn.Module):

600

601 self.config = config

602 self.quant_config = quant_config

603 - self.padding_idx = config.pad_token_id

603

604 self.vocab_size = config.vocab_size

605

vllm/model_executor/models/jais2.py

-1

@@ -305,7 +305,6 @@ class Jais2Model(nn.Module):

305

306 self.config = config

307 self.quant_config = quant_config

308 - self.padding_idx = config.pad_token_id

308

309 self.vocab_size = config.vocab_size

310 self.org_vocab_size = config.vocab_size

vllm/model_executor/models/kimi_linear.py

-1

@@ -393,7 +393,6 @@ class KimiLinearModel(nn.Module):

393 parallel_config = vllm_config.parallel_config

394 self.config = config

395

396 - self.padding_idx = config.pad_token_id

396 self.vocab_size = config.vocab_size

397

398 if get_pp_group().is_first_rank:

vllm/model_executor/models/longcat_flash.py

-1

@@ -486,7 +486,6 @@ class FlashModel(nn.Module):

486 quant_config = vllm_config.quant_config

487 self.config = config

488

489 - self.padding_idx = getattr(config, "pad_token_id", None)

489 self.vocab_size = config.vocab_size

490

491 if get_pp_group().is_first_rank:

vllm/model_executor/models/minimax_text_01.py

-1

@@ -495,7 +495,6 @@ class MiniMaxText01Model(nn.Module):

495 cache_config = vllm_config.cache_config

496 scheduler_config = vllm_config.scheduler_config

497

498 - self.padding_idx = config.pad_token_id

498 self.vocab_size = config.vocab_size

499

500 self.decoder_attention_types = getattr(

vllm/model_executor/models/nemotron_nas.py

-1

@@ -241,7 +241,6 @@ class DeciModel(nn.Module):

241

242 self.config = config

243 self.quant_config = quant_config

244 - self.padding_idx = config.pad_token_id

244

245 self.vocab_size = config.vocab_size

246

vllm/model_executor/models/openpangu.py

-1

@@ -1029,7 +1029,6 @@ class OpenPanguModel(nn.Module):

1029 self.config = config

1030 self.num_redundant_experts = eplb_config.num_redundant_experts

1031

1032 - self.padding_idx = config.pad_token_id

1032 self.vocab_size = config.vocab_size

1033

1034 if get_pp_group().is_first_rank or (

vllm/model_executor/models/plamo2.py

-1

@@ -748,7 +748,6 @@ class Plamo2Model(torch.nn.Module):

748 config = vllm_config.model_config.hf_config

749

750 self.config = config

751 - self.padding_idx = config.pad_token_id

751 self.vocab_size = config.vocab_size

752

753 self.embed_tokens = VocabParallelEmbedding(

vllm/model_executor/models/plamo3.py

-1

@@ -317,7 +317,6 @@ class Plamo3Model(nn.Module):

317 config = vllm_config.model_config.hf_config

318

319 self.config = config

320 - self.padding_idx = config.pad_token_id

320 self.vocab_size = config.vocab_size

321 self.org_vocab_size = config.vocab_size

322

vllm/model_executor/models/qwen3_moe.py

-1

@@ -443,7 +443,6 @@ class Qwen3MoeModel(nn.Module):

443 eplb_config = parallel_config.eplb_config

444 self.num_redundant_experts = eplb_config.num_redundant_experts

445

446 - self.padding_idx = config.pad_token_id

446 self.vocab_size = config.vocab_size

447 self.config = config

448 self.quant_config = quant_config