fix config image_token_id error

server/text_generation_server/models/custom_modeling/flash_pali_gemma_modeling.py

@@ -97,7 +97,6 @@ def get_inputs_embeds(
 
     def forward(
         self,
-        input_ids: torch.Tensor,
         position_ids: torch.Tensor,
         cu_seqlen_prefill: Optional[torch.Tensor],
         kv_cache: List[Tuple[torch.Tensor, torch.Tensor]],
@@ -107,12 +106,9 @@ def forward(
         max_s: int,
         prefill_cache_indices: Optional[torch.Tensor] = None,
         lm_head_indices: Optional[torch.Tensor] = None,
-        pixel_values: torch.FloatTensor = None,
         # Unused here
-        pixel_attention_mask: Optional[torch.BoolTensor] = None,
-        image_sizes: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.BoolTensor] = None,
         adapter_data: Optional[torch.Tensor] = None,
-        image_grid_thw: Optional[torch.LongTensor] = None,
         inputs_embeds: Optional[torch.Tensor] = None,
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
         # TODO This is odd but apparently pali gemma position ids start at 1.

server/text_generation_server/models/vlm_causal_lm.py

@@ -389,9 +389,8 @@ def batch_tokenized_inputs(
         max_length = 0
         vocab = tokenizer.get_vocab()
 
-        config.image_token_index = getattr(
-            config, "image_token_index", config.image_token_id
-        )
+        if not hasattr(config, "image_token_index"):
+            config.image_token_index = config.image_token_id
 
         batch_tokenized_inputs: List[List[int]] = []
         batch_image_inputs: List[Optional[List[dict]]] = []