|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- zh |
|
|
- en |
|
|
base_model: |
|
|
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B |
|
|
pipeline_tag: text-generation |
|
|
library_name: transformers |
|
|
tags: |
|
|
- Context |
|
|
- DeepSeek-R1-Distill-Qwen-1.5B |
|
|
--- |
|
|
|
|
|
# DeepSeek-R1-Distill-Qwen-1.5B |
|
|
|
|
|
This version of DeepSeek-R1-Distill-Qwen-1.5B has been converted to run on the Axera NPU using **w8a16** and **w4a16** quantization. |
|
|
|
|
|
This model has been optimized with the following LoRA: |
|
|
|
|
|
Compatible with Pulsar2 version: 4.2 |
|
|
|
|
|
## Feature |
|
|
|
|
|
- Support for longer contexts, in this sample it's 2k |
|
|
- Support context dialogue |
|
|
- System prompt kvcache is supported |
|
|
|
|
|
## Convert tools links: |
|
|
|
|
|
For those who are interested in model conversion, you can try to export axmodel through the original repo : https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B and https://huggingface.co/jakiAJK/DeepSeek-R1-Distill-Qwen-1.5B_GPTQ-int4 |
|
|
|
|
|
[Pulsar2 Link, How to Convert LLM from Huggingface to axmodel](https://pulsar2-docs.readthedocs.io/en/latest/appendix/build_llm.html) |
|
|
|
|
|
[AXera NPU AXEngine LLM Runtime](https://github.com/AXERA-TECH/ax-llm/tree/ax-context) |
|
|
|
|
|
[AXera NPU AXCL LLM Runtime](https://github.com/AXERA-TECH/ax-llm/tree/axcl-context) |
|
|
|
|
|
### Convert script |
|
|
|
|
|
The follow show how to convert DeepSeek-R1-Distill-Qwen-1.5B |
|
|
|
|
|
``` |
|
|
pulsar2 llm_build --input_path deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B \ |
|
|
--output_path deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B-ax650 \ |
|
|
--hidden_state_type bf16 --kv_cache_len 2047 --prefill_len 128 \ |
|
|
--last_kv_cache_len 128 \ |
|
|
--last_kv_cache_len 256 \ |
|
|
--last_kv_cache_len 384 \ |
|
|
--last_kv_cache_len 512 \ |
|
|
--last_kv_cache_len 640 \ |
|
|
--last_kv_cache_len 768 \ |
|
|
--last_kv_cache_len 896 \ |
|
|
--last_kv_cache_len 1024 \ |
|
|
--last_kv_cache_len 1152 \ |
|
|
--last_kv_cache_len 1280 \ |
|
|
--last_kv_cache_len 1408 \ |
|
|
--last_cache_len 1536 \ |
|
|
--chip AX650 -c 1 --parallel 8 |
|
|
``` |
|
|
|
|
|
## Support Platform |
|
|
|
|
|
- AX650 |
|
|
- AX650N DEMO Board |
|
|
- [M4N-Dock(爱芯派Pro)](https://wiki.sipeed.com/hardware/zh/maixIV/m4ndock/m4ndock.html) |
|
|
- [M.2 Accelerator card](https://axcl-docs.readthedocs.io/zh-cn/latest/doc_guide_hardware.html) |
|
|
- AX630C |
|
|
- *TBD* |
|
|
|
|
|
|Chips|w8a16|w4a16| DDR | Flash | |
|
|
|--|--|--|--|--| |
|
|
|AX650| 12 tokens/sec| 17 tokens/sec | 2.3GB | 2.3GB | |
|
|
|
|
|
## How to use |
|
|
|
|
|
Download all files from this repository to the device |
|
|
|
|
|
``` |
|
|
root@ax650:/mnt/qtang/llm-test/deepseek-r1-1.5b-ctx# tree -L 1 |
|
|
. |
|
|
|-- README.md |
|
|
|-- config.json |
|
|
|-- deepseek-r1-1.5b-ax650 |
|
|
|-- deepseek-r1-1.5b-int4-ax650 |
|
|
|-- deepseek-r1_tokenizer |
|
|
|-- deepseek-r1_tokenizer.py |
|
|
|-- figures |
|
|
|-- main_ax650 |
|
|
|-- main_axcl_aarch64 |
|
|
|-- main_axcl_x86 |
|
|
|-- post_config.json |
|
|
|-- run_deepseek-r1_1.5B_ax650.sh |
|
|
|-- run_deepseek-r1_1.5B_int4_ax650.sh |
|
|
|-- run_deepseek-r1_1.5b_axcl_aarch64.sh |
|
|
|-- run_deepseek-r1_1.5b_axcl_x86.sh |
|
|
|-- run_deepseek-r1_1.5b_int4_axcl_aarch64.sh |
|
|
`-- run_deepseek-r1_1.5b_int4_axcl_x86.sh |
|
|
|
|
|
4 directories, 16 files |
|
|
``` |
|
|
|
|
|
#### Start the Tokenizer service |
|
|
|
|
|
``` |
|
|
root@ax650:/mnt/qtang/llm-test/deepseek-r1-1.5b-ctx# python3 deepseek-r1_tokenizer_uid.py |
|
|
Server running at http://127.0.0.1:12345 |
|
|
``` |
|
|
|
|
|
#### System prompt cache |
|
|
|
|
|
- The System prompt can be preset through the configuration file from `--system_prompt` |
|
|
- The System prompt can be cached in the form of kv cache to a specified folder for quick loading at the next run time from `--kvcache_path` |
|
|
- This folder needs to be created manually before running, for example `mkdir kvcache` |
|
|
|
|
|
``` |
|
|
root@ax650:/mnt/qtang/llm-test/deepseek-r1-1.5b-ctx# cat run_deepseek-r1_1.5b_ax650.sh |
|
|
./main_ax650 \ |
|
|
--template_filename_axmodel "deepseek-r1-1.5b-ax650/qwen2_p128_l%d_together.axmodel" \ |
|
|
--axmodel_num 28 \ |
|
|
--url_tokenizer_model "http://127.0.0.1:12345" \ |
|
|
--filename_post_axmodel "deepseek-r1-1.5b-ax650/qwen2_post.axmodel" \ |
|
|
--filename_tokens_embed "deepseek-r1-1.5b-ax650/model.embed_tokens.weight.bfloat16.bin" \ |
|
|
--tokens_embed_num 151936 \ |
|
|
--tokens_embed_size 1536 \ |
|
|
--use_mmap_load_embed 1 \ |
|
|
--live_print 1 |
|
|
``` |
|
|
|
|
|
#### Inference with AX650 Host, such as M4N-Dock(爱芯派Pro) or AX650N DEMO Board |
|
|
|
|
|
Open another terminal and run `run_deepseek-r1_1.5b_ax650.sh` |
|
|
|
|
|
``` |
|
|
root@ax650:/mnt/qtang/llm-test/deepseek-r1-1.5b-ctx# ./run_deepseek-r1_1.5b_ax650.sh |
|
|
[I][ Init][ 110]: LLM init start |
|
|
[I][ Init][ 34]: connect http://127.0.0.1:12345 ok |
|
|
[I][ Init][ 57]: uid: 96cbe293-ff24-4011-b3a1-b5ab95234990 |
|
|
bos_id: 151646, eos_id: 151643 |
|
|
3% | ██ | 1 / 31 [2.07s<64.23s, 0.48 count/s] tokenizer init ok[I][ Init][ 26]: LLaMaEmbedSelector use mmap |
|
|
100% | ████████████████████████████████ | 31 / 31 [8.76s<8.76s, 3.54 count/s] init post axmodel ok,remain_cmm(8114 MB)[I][ Init][ 188]: max_token_len : 2047 |
|
|
[I][ Init][ 193]: kv_cache_size : 256, kv_cache_num: 2047 |
|
|
[I][ Init][ 201]: prefill_token_num : 128 |
|
|
[I][ Init][ 205]: grp: 1, prefill_max_token_num : 1 |
|
|
[I][ Init][ 205]: grp: 2, prefill_max_token_num : 128 |
|
|
[I][ Init][ 205]: grp: 3, prefill_max_token_num : 256 |
|
|
[I][ Init][ 205]: grp: 4, prefill_max_token_num : 384 |
|
|
[I][ Init][ 205]: grp: 5, prefill_max_token_num : 512 |
|
|
[I][ Init][ 205]: grp: 6, prefill_max_token_num : 640 |
|
|
[I][ Init][ 205]: grp: 7, prefill_max_token_num : 768 |
|
|
[I][ Init][ 205]: grp: 8, prefill_max_token_num : 896 |
|
|
[I][ Init][ 205]: grp: 9, prefill_max_token_num : 1024 |
|
|
[I][ Init][ 209]: prefill_max_token_num : 1024 |
|
|
[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][ 218]: LLM init ok |
|
|
Type "q" to exit, Ctrl+c to stop current running |
|
|
[I][ GenerateKVCachePrefill][ 275]: input token num : 11, prefill_split_num : 1 prefill_grpid : 2 |
|
|
[I][ GenerateKVCachePrefill][ 315]: input_num_token:11 |
|
|
[I][ main][ 228]: precompute_len: 11 |
|
|
[I][ main][ 229]: system_prompt: |
|
|
prompt >> 你是谁 |
|
|
[I][ SetKVCache][ 529]: prefill_grpid:2 kv_cache_num:128 precompute_len:11 input_num_token:6 |
|
|
[I][ SetKVCache][ 532]: current prefill_max_token_num:896 |
|
|
[I][ Run][ 658]: input token num : 6, prefill_split_num : 1 |
|
|
[I][ Run][ 684]: input_num_token:6 |
|
|
[I][ Run][ 807]: ttft: 256.87 ms |
|
|
<think> |
|
|
您好！我是由中国的深度求索（DeepSeek）公司开发的智能助手DeepSeek-R1。如您有任何任何问题，我会尽我所能为您提供帮助。 |
|
|
</think> |
|
|
|
|
|
您好！我是由中国的深度求索（DeepSeek）公司开发的智能助手DeepSeek-R1。如您有任何任何问题，我会尽我所能为您提供帮助。 |
|
|
|
|
|
[N][ Run][ 921]: hit eos,avg 17.68 token/s |
|
|
|
|
|
[I][ GetKVCache][ 498]: precompute_len:91, remaining:933 |
|
|
|
|
|
``` |
