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>

vllm/model_executor/models/ernie45_moe.py

@@ -421,7 +421,6 @@ class Ernie4_5_MoeModel(nn.Module):
         cache_config = vllm_config.cache_config
         quant_config = vllm_config.quant_config
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
         self.config = config
         parallel_config = vllm_config.parallel_config

vllm/model_executor/models/ernie45_vl_moe.py

@@ -523,7 +523,6 @@ class Ernie4_5_VLMoeModel(nn.Module):
         cache_config = vllm_config.cache_config
         quant_config = vllm_config.quant_config
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
         self.config = config
 

vllm/model_executor/models/granitemoeshared.py

@@ -157,7 +157,6 @@ class GraniteMoeSharedModel(nn.Module):
 
         self.config = config
         self.quant_config = quant_config  # Required by MixtralModel
-        self.padding_idx = config.pad_token_id
 
         self.vocab_size = config.vocab_size
 

vllm/model_executor/models/grok1.py

@@ -451,7 +451,6 @@ class Grok1Model(nn.Module):
 
         self.config = config
         self.quant_config = quant_config
-        self.padding_idx = config.pad_token_id
 
         # Store expert naming for weight loading
         self.ckpt_gate_proj_name = ckpt_gate_proj_name

vllm/model_executor/models/hunyuan_v1.py

@@ -600,7 +600,6 @@ class HunYuanModel(nn.Module):
 
         self.config = config
         self.quant_config = quant_config
-        self.padding_idx = config.pad_token_id
 
         self.vocab_size = config.vocab_size
 

vllm/model_executor/models/jais2.py

@@ -305,7 +305,6 @@ class Jais2Model(nn.Module):
 
         self.config = config
         self.quant_config = quant_config
-        self.padding_idx = config.pad_token_id
 
         self.vocab_size = config.vocab_size
         self.org_vocab_size = config.vocab_size

vllm/model_executor/models/kimi_linear.py

@@ -393,7 +393,6 @@ class KimiLinearModel(nn.Module):
         parallel_config = vllm_config.parallel_config
         self.config = config
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
         if get_pp_group().is_first_rank:

vllm/model_executor/models/longcat_flash.py

@@ -486,7 +486,6 @@ class FlashModel(nn.Module):
         quant_config = vllm_config.quant_config
         self.config = config
 
-        self.padding_idx = getattr(config, "pad_token_id", None)
         self.vocab_size = config.vocab_size
 
         if get_pp_group().is_first_rank:

vllm/model_executor/models/minimax_text_01.py

@@ -495,7 +495,6 @@ class MiniMaxText01Model(nn.Module):
         cache_config = vllm_config.cache_config
         scheduler_config = vllm_config.scheduler_config
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
         self.decoder_attention_types = getattr(

vllm/model_executor/models/nemotron_nas.py

@@ -241,7 +241,6 @@ class DeciModel(nn.Module):
 
         self.config = config
         self.quant_config = quant_config
-        self.padding_idx = config.pad_token_id
 
         self.vocab_size = config.vocab_size
 

vllm/model_executor/models/openpangu.py

@@ -1029,7 +1029,6 @@ class OpenPanguModel(nn.Module):
         self.config = config
         self.num_redundant_experts = eplb_config.num_redundant_experts
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
         if get_pp_group().is_first_rank or (

vllm/model_executor/models/plamo2.py

@@ -748,7 +748,6 @@ class Plamo2Model(torch.nn.Module):
         config = vllm_config.model_config.hf_config
 
         self.config = config
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
         self.embed_tokens = VocabParallelEmbedding(

vllm/model_executor/models/plamo3.py

@@ -317,7 +317,6 @@ class Plamo3Model(nn.Module):
         config = vllm_config.model_config.hf_config
 
         self.config = config
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
         self.org_vocab_size = config.vocab_size
 

vllm/model_executor/models/qwen3_moe.py

@@ -443,7 +443,6 @@ class Qwen3MoeModel(nn.Module):
         eplb_config = parallel_config.eplb_config
         self.num_redundant_experts = eplb_config.num_redundant_experts
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
         self.config = config
         self.quant_config = quant_config