# 驗測模型所需要的硬體資源

## :memo: Where do I start?
- 聯絡窗口 Email us : [e-mail] 王小姐

### 事前準備
- 映像檔下載

singularity pull c00cjz00/c00cjz00_cuda11.8_pytorch:2.1.2-cuda11.8-cudnn8-devel-llama_factory

- 原生安裝所需套件

# 套件安裝ubuntu相關套件
apt install libfontconfig  libaio-dev  libibverbs-dev jq

# 安裝 Llama-factory 相關套件
pip install llmtuner==0.5.3 deepspeed==0.13.1  bitsandbytes==0.42.0 opencc opencc-python-reimplemented

## 硬體資源需求預估

### Deepspeed Stage2, 7B

from transformers import AutoModel
from deepspeed.runtime.zero.stage_1_and_2 import estimate_zero2_model_states_mem_needs_all_live
model = AutoModel.from_pretrained("/work/u00cjz00/slurm_jobs/github/models/Llama-2-7b-chat-hf")
estimate_zero2_model_states_mem_needs_all_live(model, num_gpus_per_node=1, num_nodes=1)
estimate_zero2_model_states_mem_needs_all_live(model, num_gpus_per_node=2, num_nodes=1)
estimate_zero2_model_states_mem_needs_all_live(model, num_gpus_per_node=4, num_nodes=1)
estimate_zero2_model_states_mem_needs_all_live(model, num_gpus_per_node=8, num_nodes=1)
estimate_zero2_model_states_mem_needs_all_live(model, num_gpus_per_node=8, num_nodes=2)


### Deepspeed Stage3, 7B

from transformers import AutoModel
from deepspeed.runtime.zero.stage3 import estimate_zero3_model_states_mem_needs_all_live
model = AutoModel.from_pretrained("/work/u00cjz00/slurm_jobs/github/models/Llama-2-7b-chat-hf")
estimate_zero3_model_states_mem_needs_all_live(model, num_gpus_per_node=1, num_nodes=1)
estimate_zero3_model_states_mem_needs_all_live(model, num_gpus_per_node=1, num_nodes=2)
#estimate_zero3_model_states_mem_needs_all_live(model, num_gpus_per_node=1, num_nodes=4)
#estimate_zero3_model_states_mem_needs_all_live(model, num_gpus_per_node=1, num_nodes=8)
#estimate_zero3_model_states_mem_needs_all_live(model, num_gpus_per_node=4, num_nodes=2)
#estimate_zero3_model_states_mem_needs_all_live(model, num_gpus_per_node=8, num_nodes=2)