codellama fine-tune 01

不能识别设备

添加依赖:export CUDA_VISIBLE_DEVICES=0,1

bitsandbytes库报错：libbitsandbytes_cpu.so: undefined symbol: cget_col_row_stats

修改bitsandbytes（直接用gpu.so替换cpu.so）
cd /home/xxx/.conda/envs/xxx/lib/python3.x/site-packages/bitsandbytes
cp libbitsandbytes_cuda1xx.so libbitsandbytes_cpu.so

如何查看torch_cuda：

import torch

print(f'Torch CUDA version: {torch.version.cuda}')

AttributeError: ‘DeepSpeedCPUAdam’ object has no attribute ‘ds_opt_adam’

解决办法： You can try this by installing deepspeed w/o any DS_* variables or via: “DS_BUILD_OPS=0 pip install deepspeed”.

After install you can force a build of cpu Adam in a Python shell via:

import deepspeed
deepspeed.ops.op_builder.CPUAdamBuilder().load()

You’ll need ninja installed for this to work, many setups already have this though.

import deepspeed

deepspeed.ops.op_builder.CPUAdamBuilder().load()

'sudo apt-get install build-essential’下不了gcc、g++

使用aptitude包依赖管理工具代替apt来处理，aptitude软件包管理工具在解决依赖性问题上更有优势，具体使用方法如下：
sudo apt-get install aptitude
sudo aptitude install gcc g++