6.本地安装Fingpt

文章目录

- [1 依赖环境设置](#1 依赖环境设置)
- [2.DeepSeek R1模型部署](#2.DeepSeek R1模型部署)
- 3.FinGPT模型部署
- 4.总结与思考

1 依赖环境设置

本地环境

Python 3.10
CUbDA (可选)
PyTorch

云环境-魔塔

一键搭建https://www.modelscope.cn/
·
预装CUbDA
PyTorch运行环境
提供GPU免费试用时长
pip install modelscope

2.DeepSeek R1模型部署

DeepSeek R1

1.5B
•CPU: 4核心
•内存: 8G
•显卡：非必须
•场景：聊天机器人，简单对话
7B/8B
•CPU: 6-8核心
•内存: 16G
•显卡：推荐8GB显存，例如RTX 3060/4060
•场景：文本摘要，翻译，轻量级多轮对话
14B
•CPU: 12核心
•内存: 64G
•显卡：推荐16GB显存，例如RTX 3080/4080
•场景：复杂任务，长文本理解与生成
32B
•CPU: 16核心
•内存: 128G
•显卡：推荐24GB显存，例如RTX 3090/4090, A100或V100
•场景：高精度专业任务，医疗/金融/法律咨询

shell 复制代码

Linux环境执行如下命令(官方)
curl-fsSL https://ollama.com/install.sh|sh
魔塔环境
modelscope download--model=modelscope/ollama-linux--local _ dir./ollama
cd ollama-linux
sudo chmod+x./ollama-modelscopeinstall.sh
./ollama-modelscopeinstall.sh
ollama serve
环境参数
OLLAMA_MODELS=自定义路径
多GPU支持
CUDA_VISIBLE_DEVICES=0,1
通过ollama命令拉取模型
ollama run deepseek-r1:7b

3.FinGPT模型部署

安装FinGPT Python依赖包

shell 复制代码

pip install-Utransformers==4.40.1peft==0.5.0
pip install-U sentencepiece
pip install-U accelerate
pip install-U datasets
pip install-U bitsandbytes
如果模型在CPU上运行
pip3 install torch torchvision torchaudio
如果模型在GPU上运行
#确定CUDA版本nvcc- version
输出： Cuda compilation tools, release 12.1, V12.1.66
pip3 install torch torchvision torchaudio-index-url https://download. pytorch. org/ whl/ cu121

验证pytorch安装结果

python 复制代码

import torch
print(torch._version_)
print(torch.cuda.is_available())

运行FinGPT

python 复制代码

import torch
from transformers import  LlamaForCausalLM, LlamaTokenizerFast
from peft import PeftModel  # 0.5.0


base_model = "meta-llama/Meta-Llama-3-8B"
peft_model = "FinGPT/fingpt-mt_llama3-8b_lora"
prompt = [
    """Instruction: What is the sentiment of this news? Please choose an answer from {negative/neutral/positive}
    Input: FINANCING OF ASPOCOMP 'S GROWTH Aspocomp is aggressively pursuing its growth strategy by increasingly focusing on technologically more demanding HDI printed circuit boards PCBs.
    Answer:""",
    """Instruction: What is the sentiment of this news? Please choose an answer from {negative/neutral/positive}
    Input: According to Gran , the company has no plans to move all production to Russia , although that is where the company is growing.
    Answer:"""
]

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")



model = LlamaForCausalLM.from_pretrained(base_model, trust_remote_code=True, device_map="cuda", torch_dtype=torch.float16)
model = PeftModel.from_pretrained(model, peft_model)
model = model.eval()
model = model.to(device)

tokenizer = LlamaTokenizerFast.from_pretrained(base_model, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
tokens = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True).to(device)

res =model.generate(**tokens, max_new_tokens=512)
res_sentences = [tokenizer.decode(i) for i in res]
out_text=[o.split("Answer:")[1] for o in res_sentences]


for sentiment in out_text:
    print(sentiment)

4.总结与思考

如何通过Ollama部署DeepSeek R1模型？
FinGPT不同版本之间的差异是什么？
FinGPT使用哪种技术调优基础模型？
如何设计Prompt指导FinGPT完成情感分析任务？