Upgrade transformers version to 4.36.0 (#2046)

requirements-rocm.txt

@@ -10,7 +10,7 @@ numpy
 tokenizers>=0.15.0
 huggingface_hub<0.18,>=0.16.4
 einops  # Required for phi-1_5
-transformers >= 4.34.0  # Required for Mistral.
+transformers >= 4.36.0  # Required for Mixtral.
 fastapi
 uvicorn[standard]
 pydantic == 1.10.13  # Required for OpenAI server.

requirements.txt

@@ -7,7 +7,7 @@ sentencepiece  # Required for LLaMA tokenizer.
 numpy
 einops  # Required for phi-1_5
 torch >= 2.1.1
-transformers >= 4.34.0  # Required for Mistral.
+transformers >= 4.36.0  # Required for Mixtral.
 xformers >= 0.0.23  # Required for CUDA 12.1.
 fastapi
 uvicorn[standard]

vllm/model_executor/models/mixtral.py

@@ -29,7 +29,7 @@ import torch
 import torch.nn.functional as F
 
 from torch import nn
-from transformers import MistralConfig
+from transformers import MixtralConfig
 
 try:
     import megablocks.ops as ops
@@ -395,7 +395,7 @@ class MixtralDecoderLayer(nn.Module):
 
     def __init__(
         self,
-        config: MistralConfig,
+        config: MixtralConfig,
     ) -> None:
         super().__init__()
         self.hidden_size = config.hidden_size
@@ -443,7 +443,7 @@ class MixtralForCausalLM(nn.Module):
 
     def __init__(
         self,
-        config: MistralConfig,
+        config: MixtralConfig,
         linear_method: Optional[LinearMethodBase] = None,
     ) -> None:
         super().__init__()