upgrade vllm from 0.6.2 to the latest version

aria/vllm/aria.py

@@ -27,13 +27,14 @@
 from transformers.utils import logging
 from vllm.attention import AttentionMetadata
 from vllm.config import CacheConfig, LoRAConfig, MultiModalConfig
+from vllm.config import VllmConfig
 from vllm.distributed import (
     get_pp_group,
     get_tensor_model_parallel_rank,
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.inputs import INPUT_REGISTRY, LLMInputs
+from vllm.inputs import INPUT_REGISTRY, LLMInputs, token_inputs
 from vllm.model_executor.layers.fused_moe import fused_moe
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
@@ -337,6 +338,10 @@ def __init__(
         prefix: str = "",
     ) -> None:
         nn.Module.__init__(self)
+
+        # FIXME(zhoufan): this is a hack to avoid the error: AttributeError: 'AriaMoELMModel' object has no attribute 'do_not_compile'. 
+        self.do_not_compile = True
+
         self.config = config
         self.padding_idx = config.pad_token_id
         lora_vocab = (
@@ -679,19 +684,15 @@ def input_processor(ctx, llm_inputs):
     # TODO: Supports dynamic image size support
     setattr(model_config.multimodal_config, "max_image_size", max(max_image_size))
 
-    new_prompt, new_token_ids = repeat_and_pad_placeholder_tokens(
+    new_prompt, new_token_ids, ranges = repeat_and_pad_placeholder_tokens(
         tokenizer,
         llm_inputs.get("prompt"),
         llm_inputs["prompt_token_ids"],
         placeholder_token_id=hf_config.image_token_index,
         repeat_count=image_feature_sizes,
     )
 
-    return LLMInputs(
-        prompt=new_prompt,
-        prompt_token_ids=new_token_ids,
-        multi_modal_data=multi_modal_data,
-    )
+    return token_inputs(prompt_token_ids=new_token_ids, prompt=new_prompt, multi_modal_data=multi_modal_data, multi_modal_placeholders={"image": ranges})
 
 
 # adapted from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration
@@ -708,12 +709,18 @@ class AriaForConditionalGeneration(nn.Module, SupportsMultiModal):
 
     def __init__(
         self,
-        config: AriaConfig,
-        multimodal_config: MultiModalConfig,
-        cache_config: Optional[CacheConfig] = None,
-        quant_config: Optional[QuantizationConfig] = None,
+        vllm_config: VllmConfig,
+        prefix: str = "",
+        # config: AriaConfig,
+        # multimodal_config: MultiModalConfig,
+        # cache_config: Optional[CacheConfig] = None,
+        # quant_config: Optional[QuantizationConfig] = None,
     ):
         super().__init__()
+        config = vllm_config.model_config.hf_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+
         # prepare the image_size to tokens mapping for the image preprocess, see input_processor
         setattr(
             config,

pyproject.toml

@@ -7,13 +7,13 @@ authors = [
 readme = "README.md"
 requires-python = ">=3.10"
 dependencies = [
-    "torch==2.4.0",
-    "torchvision==0.19.0",
+    "torch==2.5.1",
+    "torchvision==0.20.1",
     "accelerate==0.34.1",
     "deepspeed==0.15.0",
     "peft==0.12.0",
     "sentencepiece==0.2.0",
-    "transformers==4.45.0",
+    "transformers==4.46.3",
     "trl==0.9.6",
     "pillow==10.4.0",
     "wandb==0.18.1",
@@ -32,7 +32,7 @@ dev = [
     "pytest==8.3.3",
 ]
 vllm = [
-    "vllm==0.6.2"
+    "vllm==0.6.4.post1"
 ]
 grouped_gemm = [
     "grouped_gemm==0.1.6"