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

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

vllm/model_executor/models/arctic.py

@@ -370,7 +370,6 @@ class ArcticModel(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(
             self.vocab_size,

vllm/model_executor/models/baichuan.py

@@ -267,7 +267,6 @@ class BaiChuanModel(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(

vllm/model_executor/models/bart.py

@@ -725,7 +725,6 @@ class BartModel(nn.Module):
 
         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/chameleon.py

@@ -851,7 +851,6 @@ class ChameleonModel(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(
             self.vocab_size,

vllm/model_executor/models/deepseek.py

@@ -339,7 +339,6 @@ class DeepseekModel(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/deepseek_v2.py

@@ -570,7 +570,6 @@ class DeepseekV2Model(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
 
         if get_pp_group().is_first_rank:

vllm/model_executor/models/exaone.py

@@ -313,7 +313,6 @@ class ExaoneModel(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/florence2.py

@@ -592,7 +592,6 @@ class Florence2LanguageModel(nn.Module):
 
         self.config = config
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
 
         self.shared = BartScaledWordEmbedding(self.vocab_size, config.d_model)

vllm/model_executor/models/fuyu.py

@@ -255,7 +255,6 @@ class FuyuForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
         self.config = config
         self.multimodal_config = multimodal_config
 
-        self.padding_idx = config.pad_token_id
         self.vocab_size = config.text_config.vocab_size
         self.image_token_id = _IMAGE_TOKEN_ID
         self.image_feature_size = config.patch_size**2 * config.num_channels

vllm/model_executor/models/granite.py

@@ -260,7 +260,6 @@ class GraniteModel(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/granitemoe.py

@@ -252,7 +252,6 @@ class GraniteMoeModel(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/idefics3.py

@@ -404,7 +404,6 @@ class Idefics3Model(nn.Module):
         quant_config = vllm_config.quant_config
 
         self.config = config
-        self.padding_idx = self.config.text_config.pad_token_id
         self.vocab_size = self.config.text_config.vocab_size
         self.vision_model = Idefics3VisionTransformer(
             config.vision_config,

vllm/model_executor/models/internlm2.py

@@ -261,7 +261,6 @@ class InternLM2Model(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.tok_embeddings = VocabParallelEmbedding(
             config.vocab_size,

vllm/model_executor/models/jamba.py

@@ -271,7 +271,6 @@ class JambaModel(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/llama.py

@@ -302,7 +302,6 @@ class LlamaModel(nn.Module):
 
         self.config = config
         self.quant_config = quant_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/mamba.py

@@ -90,7 +90,6 @@ class MambaModel(nn.Module):
         is_lora_enabled = bool(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/minicpm.py

@@ -365,7 +365,6 @@ class MiniCPMModel(nn.Module):
         self.config = config
         self.cache_config = cache_config
         self.quant_config = quant_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/mixtral.py

@@ -254,7 +254,6 @@ class MixtralModel(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/mixtral_quant.py

@@ -302,7 +302,6 @@ class MixtralModel(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/mllama.py

@@ -1031,7 +1031,6 @@ class MllamaTextModel(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(config.vocab_size + 8,
                                                    config.hidden_size)

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,
+    def __init__(self, num_positions: int, embedding_dim: int):
-                 num_positions: int,
-                 embedding_dim: int,
-                 padding_idx: Optional[int] = None):
         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)