Bump transformers version for Llama 3.1 hotfix and patch Chameleon (#6690)

vllm/model_executor/models/__init__.py

@@ -16,8 +16,6 @@ _GENERATION_MODELS = {
     "BaiChuanForCausalLM": ("baichuan", "BaiChuanForCausalLM"),  # baichuan-7b
     "BaichuanForCausalLM": ("baichuan", "BaichuanForCausalLM"),  # baichuan-13b
     "BloomForCausalLM": ("bloom", "BloomForCausalLM"),
-    #TODO(ywang96): remove this when huggingface fixes the model repo
-    "ChameleonForCausalLM": ("chameleon", "ChameleonForConditionalGeneration"),
     "ChameleonForConditionalGeneration":
     ("chameleon", "ChameleonForConditionalGeneration"),
     "ChatGLMModel": ("chatglm", "ChatGLMForCausalLM"),

vllm/model_executor/models/chameleon.py

@@ -6,6 +6,7 @@ import torch
 import torch.nn.functional as F
 from PIL import Image
 from torch import nn
+from transformers import ChameleonConfig, ChameleonVQVAEConfig
 
 from vllm.attention import Attention, AttentionMetadata
 from vllm.config import CacheConfig, MultiModalConfig
@@ -30,8 +31,6 @@ from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.image import (cached_get_tokenizer,
                                    repeat_and_pad_image_tokens)
 from vllm.sequence import IntermediateTensors, SamplerOutput, SequenceData
-from vllm.transformers_utils.configs import (ChameleonConfig,
-                                             ChameleonVQVAEConfig)
 from vllm.utils import print_warning_once
 
 from .interfaces import SupportsVision