跳到主要内容

DeepSeek-R1-Qwen-1.5B

此文档讲解如何在安装了瑞莎智核 AX-M1 的 host 设备上运行 DeepSeek-R1-Qwen-1.5B 示例应用。

预编译模型量化方式：w4a16 和 w8a16

下载示例应用仓库​

使用 huggingfcae-cli 下载示例应用仓库。

Host
pip3 install -U "huggingface_hub[cli]"
huggingface-cli download AXERA-TECH/DeepSeek-R1-Distill-Qwen-1.5B --local-dir ./DeepSeek-R1-Distill-Qwen-1.5B
cd DeepSeek-R1-Distill-Qwen-1.5B
Host
pip3 install -U "huggingface_hub[cli]"
huggingface-cli download AXERA-TECH/DeepSeek-R1-Distill-Qwen-1.5B --local-dir ./DeepSeek-R1-Distill-Qwen-1.5B
cd DeepSeek-R1-Distill-Qwen-1.5B

示例使用​

安装 python 依赖​

Host
pip3 install transformers==4.53.3 jinja2==3.1.6

启动 Tokenizer 服务​

Host
python3 deepseek-r1_tokenizer.py --port 12345 > /dev/null 2>&1 &
(.venv) rock@rock-5b-plus:~/ssd/axera/DeepSeek-R1-Distill-Qwen-1.5B$ python3 deepseek-r1_tokenizer.py
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
151646 <｜begin▁of▁sentence｜> 151643 <｜end▁of▁sentence｜>
<｜begin▁of▁sentence｜>You are Qwen, created by Alibaba Cloud. You are a helpful assistant.<｜User｜>hello world<｜Assistant｜>
[151646, 151646, 2610, 525, 1207, 16948, 11, 3465, 553, 54364, 14817, 13, 1446, 525, 264, 10950, 17847, 13, 151644, 14990, 1879, 151645]
http://localhost:12345
提示

如需结束后台的 Tokenizer 服务，请使用 jobs 查看后台编号，然后使用 kill %N 结束后台进程, 这里的 %N 是 jobs 下的后台编号

模型推理​

Host
chmod +x main_axcl_aarch64
mkdir kvcache
bash run_deepseek-r1_1.5b_axcl_aarch64.sh
提示

请检查 run_xxx.sh 运行脚本中 tokenizer_model 的端口是否与 Tokenizer 服务端口一致

rock@rock-5b-plus:~/ssd/axera/DeepSeek-R1-Distill-Qwen-1.5B$ bash run_deepseek-r1_1.5b_axcl_aarch64.sh
build time: Feb 13 2025 15:44:57
[I][                            Init][ 111]: LLM init start
bos_id: 151646, eos_id: 151643
100% | ████████████████████████████████ |  31 /  31 [26.50s<26.50s, 1.17 count/s] init post axmodel okremain_cmm(-1 MB)
[I][                            Init][ 226]: max_token_len : 1023
[I][                            Init][ 231]: kv_cache_size : 256, kv_cache_num: 1023
[I][                     load_config][ 282]: load config:
{
    "enable_repetition_penalty": false,
    "enable_temperature": true,
    "enable_top_k_sampling": true,
    "enable_top_p_sampling": false,
    "penalty_window": 20,
    "repetition_penalty": 1.2,
    "temperature": 0.9,
    "top_k": 10,
    "top_p": 0.8
}

[I][                            Init][ 288]: LLM init ok
Type "q" to exit, Ctrl+c to stop current running

性能参考​

模型量化方式host 设备token/s
DeepSeek-R1-Distill-Qwen-1.5Bw8a16ROCK 5B+8.95
DeepSeek-R1-Distill-Qwen-1.5B-GPTQ-Int4w4a16ROCK 5B+12.65