InternVL3-1B

This version of InternVL3-1B 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: 4.1

Convert tools links:

For those who are interested in model conversion, you can try to export axmodel through the original repo : https://huggingface.co/OpenGVLab/InternVL3-1B

How to Convert LLM from Huggingface to axmodel

AXera NPU HOST LLM Runtime

AXera NPU AXCL LLM Runtime

Support Platform

• AX650
  • AX650N DEMO Board
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card

Chips	image encoder 448	ttft	w8a16
AX650	380 ms	623 ms	30 tokens/sec

How to use

Download all files from this repository to the device

root@ax650:/mnt/qtang/llm-test/internvl3-1b# tree -L 1
.
|-- gradio_demo.py
|-- internvl3_1b_ax650
|-- internvl3_tokenizer
|-- internvl3_tokenizer.py
|-- main_api_ax650
|-- main_api_axcl_x86
|-- main_ax650
|-- main_axcl_x86
|-- post_config.json
|-- run_internvl_3_1b_448_api_ax650.sh
|-- run_internvl_3_1b_448_api_axcl_x86.sh
|-- run_internvl_3_1b_448_ax650.sh
|-- run_internvl_3_1b_448_axcl_x86.sh
`-- ssd_car.jpg

Install transformer

pip install transformers==4.41.1

Start the Tokenizer service

root@ax650:/mnt/qtang/llm-test/internvl3-1b# python3 internvl3_tokenizer.py
None None 151645 <|im_end|> 151665 151667
context_len is  256
prompt is <|im_start|>system
你是书生·万象, 英文名是InternVL, 是由上海人工智能实验室、清华大学及多家合作单位联合开发的多模态大语言模型.<|im_end|>
......
http://0.0.0.0:12345

Inference with AX650 Host, such as M4N-Dock(爱芯派Pro) or AX650 DEMO Board

• input text

描述下图片

• input image

Open another terminal and run ./run_internvl3_1b_448_ax650.sh

root@ax650:/mnt/qtang/llm-test/internvl3-1b# ./run_internvl_3_1b_448_ax650.sh
[I][                            Init][ 134]: LLM init start
[I][                            Init][  34]: connect http://0.0.0.0:12345 ok
bos_id: -1, eos_id: 151645
img_start_token: 151665
img_context_token: 151667
  3% | ██                                |   1 /  27 [0.01s<0.32s, 83.33 count/s] tokenizer init ok
[I][                            Init][  45]: LLaMaEmbedSelector use mmap
  7% | ███                               |   2 /  27 [0.01s<0.19s, 142.86 count/s] embed_selector init ok
100% | ████████████████████████████████ |  27 /  27 [6.92s<6.92s, 3.90 count/s] init post axmodel ok,remain_cmm(11068 MB)
[I][                            Init][ 226]: IMAGE_CONTEXT_TOKEN: 151667, IMAGE_START_TOKEN: 151665
[I][                            Init][ 251]: image encoder input nchw@float32
[I][                            Init][ 281]: image encoder output float32
[I][                            Init][ 291]: image_encoder_height : 448, image_encoder_width: 448
[I][                            Init][ 293]: max_token_len : 2047
[I][                            Init][ 296]: kv_cache_size : 128, kv_cache_num: 2047
[I][                            Init][ 304]: prefill_token_num : 128
[I][                            Init][ 308]: grp: 1, prefill_max_token_num : 1
[I][                            Init][ 308]: grp: 2, prefill_max_token_num : 128
[I][                            Init][ 308]: grp: 3, prefill_max_token_num : 256
[I][                            Init][ 308]: grp: 4, prefill_max_token_num : 384
[I][                            Init][ 308]: grp: 5, prefill_max_token_num : 512
[I][                            Init][ 308]: grp: 6, prefill_max_token_num : 640
[I][                            Init][ 308]: grp: 7, prefill_max_token_num : 768
[I][                            Init][ 308]: grp: 8, prefill_max_token_num : 896
[I][                            Init][ 308]: grp: 9, prefill_max_token_num : 1024
[I][                            Init][ 312]: 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][ 321]: LLM init ok
Type "q" to exit, Ctrl+c to stop current running
prompt >> 描述下图片
image >> ssd_car.jpg
[I][                          Encode][ 415]: image encode time : 387.35 ms, size : 229376
[I][                          Encode][ 524]: idx:0 offset : 50 out_embed.size() : 279552
[I][                             Run][ 551]: input token num : 312, prefill_split_num : 3
[I][                             Run][ 566]: prefill grpid 4
[I][                             Run][ 593]: input_num_token:128
[I][                             Run][ 593]: input_num_token:128
[I][                             Run][ 593]: input_num_token:56
[I][                             Run][ 717]: ttft: 623.71 ms
图片中出现的物体包括：

1. 一辆红色的双层巴士，巴士上有一则广告，广告上写着“THINGS GET MORE EXCITING WHEN YOU SAY YES” (当你说“是”时，事情就更兴奋了)。
2. 一位微笑的女性站在巴士旁边。
3. 一辆黑色的汽车停在路边。
4. 一家商店的橱窗。
5. 一些建筑物的外墙和窗户。
6. 一根黑色的路灯杆。

这些是图片中实际存在的物体。

[N][                             Run][ 826]: hit eos,avg 28.78 token/s

prompt >> q