Clean up unused padding_idx variables across many model definitions (#13240)

Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>

vllm/model_executor/models/nemotron.py

@@ -300,7 +300,6 @@ class NemotronModel(nn.Module):
         lora_config = vllm_config.lora_config
 
         self.config = config
-        self.padding_idx = config.pad_token_id
         lora_vocab = (lora_config.lora_extra_vocab_size *
                       (lora_config.max_loras or 1)) if lora_config else 0
         self.vocab_size = config.vocab_size + lora_vocab

vllm/model_executor/models/olmoe.py

@@ -252,7 +252,6 @@ class OlmoeModel(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.embed_tokens = VocabParallelEmbedding(

vllm/model_executor/models/opt.py

@@ -200,7 +200,6 @@ class OPTDecoder(nn.Module):
     ):
         super().__init__()
         self.config = config
-        self.padding_idx = config.pad_token_id
         self.max_target_positions = config.max_position_embeddings
         self.vocab_size = config.vocab_size
 

vllm/model_executor/models/orion.py

@@ -217,7 +217,6 @@ class OrionModel(nn.Module):
         quant_config = vllm_config.quant_config
 
         self.config = config
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
         self.embed_tokens = VocabParallelEmbedding(
             config.vocab_size,

vllm/model_executor/models/phimoe.py

@@ -441,7 +441,6 @@ class PhiMoEModel(nn.Module):
         quant_config = vllm_config.quant_config
         lora_config = vllm_config.lora_config
 
-        self.padding_idx = config.pad_token_id
         lora_vocab = ((lora_config.lora_extra_vocab_size *
                        (lora_config.max_loras or 1)) if lora_config else 0)
         self.vocab_size = config.vocab_size + lora_vocab

vllm/model_executor/models/qwen2.py

@@ -284,7 +284,6 @@ class Qwen2Model(nn.Module):
 
         self.config = config
         self.quant_config = quant_config
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
         if get_pp_group().is_first_rank or (config.tie_word_embeddings

vllm/model_executor/models/qwen2_moe.py

@@ -325,7 +325,6 @@ class Qwen2MoeModel(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.embed_tokens = VocabParallelEmbedding(

vllm/model_executor/models/solar.py

@@ -269,7 +269,6 @@ class SolarModel(nn.Module):
         lora_config = vllm_config.lora_config
 
         self.config = config
-        self.padding_idx = config.pad_token_id
         lora_vocab = ((lora_config.lora_extra_vocab_size *
                        (lora_config.max_loras or 1)) if lora_config else 0)
         self.vocab_size = config.vocab_size + lora_vocab

vllm/model_executor/models/starcoder2.py

@@ -212,10 +212,8 @@ class Starcoder2Model(nn.Module):
         quant_config = vllm_config.quant_config
 
         self.config = config
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
-        # TODO: consider padding_idx (currently removed)
         self.embed_tokens = VocabParallelEmbedding(
             config.vocab_size,
             config.hidden_size,

vllm/model_executor/models/whisper.py

@@ -49,10 +49,7 @@ class WhisperAudioInputs(TypedDict):
 
 class WhisperPositionalEmbedding(nn.Embedding):
 
-    def __init__(self,
-                 num_positions: int,
-                 embedding_dim: int,
-                 padding_idx: Optional[int] = None):
+    def __init__(self, num_positions: int, embedding_dim: int):
         super().__init__(num_positions, embedding_dim)
 
     def forward(self, position_ids):
@@ -359,7 +356,6 @@ class WhisperEncoder(nn.Module):
         config = vllm_config.model_config.hf_config
         embed_dim = config.d_model
         self.num_mel_bins = config.num_mel_bins
-        self.padding_idx = config.pad_token_id
         self.max_source_positions = config.max_source_positions
         self.embed_scale = (math.sqrt(embed_dim)
                             if config.scale_embedding else 1.0)