Qwen3
Collection
3 items
β’
Updated
Qwen3-0.6B-Int8
This version of Qwen3-0.6B-Int8 has been converted to run on the Axera NPU using w8a16 quantization.
This model has been optimized with the following LoRA:
Compatible with Pulsar2 version: 4.0-temp(Not released yet)
Convert tools links:
For those who are interested in model conversion, you can try to export axmodel through the original repo : https://huggingface.co/Qwen/Qwen3-0.6B
Pulsar2 Link, How to Convert LLM from Huggingface to axmodel
Support Platform
- AX650
- AX630C
- developing
| Chips | w8a16 | w4a16 |
|---|---|---|
| AX650 | 20 tokens/sec | TBD |
How to use
Download all files from this repository to the device
root@ax650:/mnt/qtang/llm-test/qwen3-0.6b# tree -L 1
.
|-- main_ax650
|-- main_axcl_aarch64
|-- main_axcl_x86
|-- post_config.json
|-- qwen2.5_tokenizer
|-- qwen3-0.6b-ax650
|-- qwen3_tokenizer
|-- qwen3_tokenizer_uid.py
|-- run_qwen3_0.6b_int8_ctx_ax650.sh
|-- run_qwen3_0.6b_int8_ctx_axcl_aarch64.sh
`-- run_qwen3_0.6b_int8_ctx_axcl_x86.sh
Start the Tokenizer service
Install requirement
pip install transformers jinja2
root@ax650:/mnt/qtang/llm-test/qwen3-0.6b# python3 qwen3_tokenizer_uid.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.
Server running at http://0.0.0.0:12345
Inference with AX650 Host, such as M4N-Dock(η±θ―ζ΄ΎPro) or AX650N DEMO Board
Open another terminal and run run_qwen3_0.6b_int8_ctx_ax650.sh
root@ax650:/mnt/qtang/llm-test/qwen3-0.6b# ./run_qwen3_0.6b_int8_ctx_ax650.sh
[I][ Init][ 110]: LLM init start
[I][ Init][ 34]: connect http://127.0.0.1:12345 ok
[I][ Init][ 57]: uid: 8199112b-da8a-4f39-ae48-9d83f422b2d3
bos_id: -1, eos_id: 151645
3% | ββ | 1 / 31 [3.76s<116.56s, 0.27 count/s] tokenizer init ok
[I][ Init][ 26]: LLaMaEmbedSelector use mmap
100% | ββββββββββββββββββββββββββββββββ | 31 / 31 [6.18s<6.18s, 5.01 count/s] init post axmodel ok,remain_cmm(10021 MB)
[I][ Init][ 188]: max_token_len : 2559
[I][ Init][ 193]: kv_cache_size : 1024, kv_cache_num: 2559
[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 : 512
[I][ Init][ 205]: grp: 3, prefill_max_token_num : 1024
[I][ Init][ 205]: grp: 4, prefill_max_token_num : 1536
[I][ Init][ 205]: grp: 5, prefill_max_token_num : 2048
[I][ Init][ 209]: prefill_max_token_num : 2048
[I][ load_config][ 282]: load config:
{
"enable_repetition_penalty": false,
"enable_temperature": false,
"enable_top_k_sampling": true,
"enable_top_p_sampling": false,
"penalty_window": 20,
"repetition_penalty": 1.2,
"temperature": 0.9,
"top_k": 1,
"top_p": 0.8
}
[I][ Init][ 218]: LLM init ok
Type "q" to exit, Ctrl+c to stop current running
[I][ GenerateKVCachePrefill][ 270]: input token num : 21, prefill_split_num : 1 prefill_grpid : 2
[I][ GenerateKVCachePrefill][ 307]: input_num_token:21
[I][ main][ 230]: precompute_len: 21
[I][ main][ 231]: system_prompt: You are Qwen, created by Alibaba Cloud. You are a helpful assistant.
prompt >> who are you
[I][ SetKVCache][ 530]: prefill_grpid:2 kv_cache_num:512 precompute_len:57 input_num_token:14
[I][ SetKVCache][ 533]: current prefill_max_token_num:1920
[I][ Run][ 659]: input token num : 14, prefill_split_num : 1
[I][ Run][ 685]: input_num_token:14
[I][ Run][ 808]: ttft: 586.92 ms
<think>
</think>
I'm Qwen, a large language model developed by Alibaba Cloud. I can help with a wide range of tasks,
from answering questions to writing code, providing information, and even assisting with creative projects.
Let me know what you need!
[N][ Run][ 922]: hit eos,avg 19.01 token/s
[I][ GetKVCache][ 499]: precompute_len:123, remaining:1925
prompt >> q
root@ax650:/mnt/qtang/llm-test/qwen3-0.6b#
Inference with M.2 Accelerator card
What is M.2 Accelerator card?, Show this DEMO based on Raspberry PI 5.
(base) axera@raspberrypi:~/samples/qwen3-0.6b $ ./run_qwen3_0.6b_int8_ctx_axcl_aarch64.sh
[I][ Init][ 136]: LLM init start
[I][ Init][ 34]: connect http://127.0.0.1:12345 ok
[I][ Init][ 57]: uid: afec8311-55c9-4785-9fed-949368362b0e
bos_id: -1, eos_id: 151645
3% | ββ | 1 / 31 [1.00s<31.12s, 1.00 count/s] tokenizer init ok
[I][ Init][ 45]: LLaMaEmbedSelector use mmap
6% | βββ | 2 / 31 [1.00s<15.56s, 1.99 count/s] embed_selector init ok
[I][ run][ 30]: AXCLWorker start with devid 0
100% | ββββββββββββββββββββββββββββββββ | 31 / 31 [28.32s<28.32s, 1.09 count/s] init post axmodel ok,remain_cmm(5068 MB)
[I][ Init][ 237]: max_token_len : 2559
[I][ Init][ 240]: kv_cache_size : 1024, kv_cache_num: 2559
[I][ Init][ 248]: prefill_token_num : 128
[I][ Init][ 252]: grp: 1, prefill_max_token_num : 1
[I][ Init][ 252]: grp: 2, prefill_max_token_num : 512
[I][ Init][ 252]: grp: 3, prefill_max_token_num : 1024
[I][ Init][ 252]: grp: 4, prefill_max_token_num : 1536
[I][ Init][ 252]: grp: 5, prefill_max_token_num : 2048
[I][ Init][ 256]: prefill_max_token_num : 2048
________________________
| ID| remain cmm(MB)|
========================
| 0| 5068|
Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―Β―
[I][ load_config][ 282]: load config:
{
"enable_repetition_penalty": false,
"enable_temperature": false,
"enable_top_k_sampling": true,
"enable_top_p_sampling": false,
"penalty_window": 20,
"repetition_penalty": 1.2,
"temperature": 0.9,
"top_k": 1,
"top_p": 0.8
}
[I][ Init][ 279]: LLM init ok
Type "q" to exit, Ctrl+c to stop current running
[I][ GenerateKVCachePrefill][ 335]: input token num : 21, prefill_split_num : 1 prefill_grpid : 2
[I][ GenerateKVCachePrefill][ 372]: input_num_token:21
[I][ main][ 236]: precompute_len: 21
[I][ main][ 237]: system_prompt: You are Qwen, created by Alibaba Cloud. You are a helpful assistant.
prompt >> who are you?
[I][ SetKVCache][ 628]: prefill_grpid:2 kv_cache_num:512 precompute_len:21 input_num_token:16
[I][ SetKVCache][ 631]: current prefill_max_token_num:1920
[I][ Run][ 869]: input token num : 16, prefill_split_num : 1
[I][ Run][ 901]: input_num_token:16
[I][ Run][1030]: ttft: 670.05 ms
<think>
</think>
I am Qwen, a large language model developed by Alibaba Cloud.
I am designed to assist with a wide range of tasks and provide helpful information.
If you have any questions or need assistance, feel free to ask!
[N][ Run][1182]: hit eos,avg 13.06 token/s
[I][ GetKVCache][ 597]: precompute_len:85, remaining:1963
prompt >> what can you do?
[I][ SetKVCache][ 628]: prefill_grpid:2 kv_cache_num:512 precompute_len:85 input_num_token:17
[I][ SetKVCache][ 631]: current prefill_max_token_num:1920
[I][ Run][ 869]: input token num : 17, prefill_split_num : 1
[I][ Run][ 901]: input_num_token:17
[I][ Run][1030]: ttft: 671.29 ms
<think>
</think>
I can help with a variety of tasks and provide assistance in different areas. For example, I can:
- Answer questions about technology, science, culture, and more.
- Help with writing, research, and problem-solving.
- Provide information and support in different languages.
- Assist with tasks such as writing, coding, and data analysis.
Let me know what you need!
[N][ Run][1182]: hit eos,avg 13.05 token/s
[I][ GetKVCache][ 597]: precompute_len:181, remaining:1867
prompt >> q
(base) axera@raspberrypi:~ $ axcl-smi
+------------------------------------------------------------------------------------------------+
| AXCL-SMI V3.4.0_20250423020139 Driver V3.4.0_20250423020139 |
+-----------------------------------------+--------------+---------------------------------------+
| Card Name Firmware | Bus-Id | Memory-Usage |
| Fan Temp Pwr:Usage/Cap | CPU NPU | CMM-Usage |
|=========================================+==============+=======================================|
| 0 AX650N V3.4.0 | 0000:01:00.0 | 182 MiB / 945 MiB |
| -- 35C -- / -- | 1% 0% | 971 MiB / 7040 MiB |
+-----------------------------------------+--------------+---------------------------------------+
+------------------------------------------------------------------------------------------------+
| Processes: |
| Card PID Process Name NPU Memory Usage |
|================================================================================================|
| 0 53261 /home/axera/samples/qwen3-0.6b/main_axcl_aarch64 953772 KiB |
+------------------------------------------------------------------------------------------------+
- Downloads last month
- 11
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
π
Ask for provider support