报错
DeepSpeed Zero-3 is not compatible with low_cpu_mem_usage=True
or with passing a `device_map
原因
DeepSpeed Zero-3的核心功能就是在显存不足的情况下,使用CPU内存
解决
如下:注释device_map=device_map行
model = transformers.AutoModelForCausalLM.from_pretrained(
model_args.model_name_or_path,
config=config,
cache_dir=training_args.cache_dir,
# device_map=device_map,# 注释
trust_remote_code=True,
quantization_config=GPTQConfig(
bits=4, disable_exllama=True
)
if training_args.use_lora and lora_args.q_lora
else None,
)