Instructions to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16
# Run inference directly in the terminal:
llama cli -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16
# Run inference directly in the terminal:
llama cli -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16
# Run inference directly in the terminal:
./build/bin/llama cli -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Use Docker

docker model run hf.co/AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Llama new
LM Studio
Jan
Ollama
How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with Ollama:
```
ollama run hf.co/AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16
```
Unsloth Desktop

How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with Docker Model Runner:
```
docker model run hf.co/AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16
```

Lemonade

How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Run and chat with the model

lemonade run user.Qwen3.5-2B-AX8860-llama.cpp-BF16

List all available models

lemonade list

Hermes Agent

How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "AXERA-TECH/Qwen3.5-2B-AX8860-llama.cpp:BF16" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Qwen3.5-2B on AX8860 (llama.cpp + ggml-axera, 2026-11-05)

在爱芯 AX8860 的 NPU（4 核）上用 llama.cpp 运行 Qwen3.5-2B，支持文本和图像输入；整个模型的 prefill 是一个 kernel、decode 是每 16 层一个 kernel，权重在 NPU 内存里只放一份。 Qwen3.5-2B with text and image input on the 4-core NPU of the Axera AX8860, with llama.cpp; the whole model's prefill is one kernel, decode one kernel per 16 layers, one copy of the weights in NPU memory.

文件 / Files

文件 / File	说明 / Description
`qwen3.5-2b-mxfp4.gguf`	模型权重（GGUF，投影层 MXFP4、embedding / 输出头 Q8_0，1.3 GB，md5 `4a494ba48e2904a7925fb4fed6cc628f`）/ weights (projections MXFP4, embedding / output head Q8_0)
`mmproj-Qwen3.5-2B-BF16.gguf`	视觉编码器权重（BF16，0.8 GB，md5 `47c5b21dc254ba4f0d4bd34454442e31`）/ vision encoder weights
`bin/`	llama.cpp 板端程序（静态链接，只依赖系统 `/opt/lib` 与 libc）/ the board programs (static; need only the system `/opt/lib` and libc)
`kernels/`	NPU kernel：prefill（整模型）、decode（16 层一组）、ViT、输出头 + `manifest.json`（上下文 16384）/ prefill (whole model), decode (16-layer groups), ViT, output head + `manifest.json` (context 16384)
`run.sh`	运行脚本 / run script

权重只来自 GGUF：kernel 不含权重，加载时从 GGUF 转换到 NPU（2.7 s）；NPU 内存里只放一份权重（prefill 和 decode 共用）。 The weights come only from the GGUFs: the kernels hold none; they are converted from the GGUF at load (2.7 s); one copy of the weights in NPU memory (shared by prefill and decode).

运行 / Run

环境 / Requirements：AX8860，SDK V0.5.0 或更新（cat /proc/ax_proc/version）/ SDK V0.5.0 or newer.

./run.sh "The capital of France is" 64                 # 文本补全 / text completion
./run.sh -i bus.jpg "Describe this image." 256         # 图像描述 / describe an image
./run.sh -c                                            # 命令行多轮对话 / multi-turn chat in the terminal
./run.sh -s 8080                                       # OpenAI 兼容服务 / OpenAI-compatible server
curl http://<board>:8080/v1/chat/completions -H 'Content-Type: application/json' \
  -d '{"messages":[{"role":"user","content":"你好"}]}'

直接调用程序时设置 / When calling the programs directly: LD_LIBRARY_PATH=/opt/lib GGML_AXERA_KERNEL_DIR=<repo>/kernels，-ngl 99，上下文 -c 最大 16384，服务端 -np 1；prompt 用 -ub 4096 -b 4096 最快 / -ub 4096 -b 4096 for the fastest prompt processing。GGML_AXERA_PROF=1 退出时打印各阶段耗时 / prints per-stage times at exit

性能 / Performance

AX8860，SDK V0.5.1，NPU 4 核 / 4 cores：

项 / Item	结果 / Result
4k prompt（4000 token）prefill	16000 tokens/s（每 256 token 一页 15 ms；首次 prompt 另加 2.7 s 权重转换 / plus a one-time 2.7 s weight conversion on the first prompt）
decode（短上下文 / short context）	121.5 tokens/s
图像编码 / image encoding	204 ms（整个 ViT 一个 kernel / the whole ViT in one kernel）+ 20 ms CPU 预处理 / preprocessing；输入缩放到约 1024 x 768（最多 768 个图像 token = 3072 个 16 x 16 patch，`--image-max-tokens 768`；更大的图缩小、更小的图放大到这个面积）/ the image is scaled to ~1024 x 768 (up to 768 image tokens = 3072 patches of 16 x 16; larger images are scaled down, smaller ones up)
加载 / load	5.5 s（热）/ 30 s（冷）
NPU 内存（CMM）/ NPU memory	加载后 / after load 2.05 GB，峰值 / peak 2.05 GB
系统内存峰值 / host memory peak	0.8 GB

正确性 / Correctness

"The capital of France is" → "Paris. …"；图像描述正确（车辆、文字、人物）/ image descriptions correct.
中文 PPL（内部测试集，-c 512）：NPU 38.85，同一 GGUF 在 CPU 上 38.00（+2.2%）/ Chinese PPL on an internal set: NPU 38.85 vs 38.00 for the same GGUF on the CPU (+2.2%).
同一输入多次运行，结果偶有极小差异（NPU 上累加顺序不固定），不影响输出质量。 Repeated runs of one input may differ slightly (the NPU's accumulation order varies); the output quality is the same.

已知限制 / Known limits

运行时独占 NPU 4 核，同时运行其他 NPU 程序会出错 / takes all 4 NPU cores; do not run other NPU programs alongside.
单序列：服务端须 -np 1 / one sequence: the server needs -np 1.
上下文最大 16384（长上下文版本随后）/ context up to 16384 (longer contexts to follow).
没有投机解码（这个尺寸的模型没有 MTP 头）/ no speculative decoding (no MTP head at this size).

源码 / Sources

llama.cpp：ZHEQIUSHUI/llama.cpp axera-dev @ 57a30bd（ggml/src/ggml-axera/models/qwen35_whole）
kernel：ZHEQIUSHUI/ax-dsl-kernels main（models/qwen35_2b/whole，kernels.md5）