@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / cb3f30c600
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 17, 05:30 UTC

Upgrade transformers version to 4.36.0 (#2046)

Woosuk Kwon committed Dec 11, 2023 at 18:39 UTC cb3f30c600169210f9715f084e34adf2afc4f7d7
3 files changed +5 -5
requirements-rocm.txt
+1 -1
@@ -10,7 +10,7 @@ numpy
10 tokenizers>=0.15.0
11 huggingface_hub<0.18,>=0.16.4
12 einops # Required for phi-1_5
13 -transformers >= 4.34.0 # Required for Mistral.
13 +transformers >= 4.36.0 # Required for Mixtral.
14 fastapi
15 uvicorn[standard]
16 pydantic == 1.10.13 # Required for OpenAI server.
requirements.txt
+1 -1
@@ -7,7 +7,7 @@ sentencepiece # Required for LLaMA tokenizer.
7 numpy
8 einops # Required for phi-1_5
9 torch >= 2.1.1
10 -transformers >= 4.34.0 # Required for Mistral.
10 +transformers >= 4.36.0 # Required for Mixtral.
11 xformers >= 0.0.23 # Required for CUDA 12.1.
12 fastapi
13 uvicorn[standard]
vllm/model_executor/models/mixtral.py
+3 -3
@@ -29,7 +29,7 @@ import torch
29 import torch.nn.functional as F
30
31 from torch import nn
32 -from transformers import MistralConfig
32 +from transformers import MixtralConfig
33
34 try:
35 import megablocks.ops as ops
@@ -395,7 +395,7 @@ class MixtralDecoderLayer(nn.Module):
395
396 def __init__(
397 self,
398 - config: MistralConfig,
398 + config: MixtralConfig,
399 ) -> None:
400 super().__init__()
401 self.hidden_size = config.hidden_size
@@ -443,7 +443,7 @@ class MixtralForCausalLM(nn.Module):
443
444 def __init__(
445 self,
446 - config: MistralConfig,
446 + config: MixtralConfig,
447 linear_method: Optional[LinearMethodBase] = None,
448 ) -> None:
449 super().__init__()