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

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Misakachain/Qwen3.5-2B 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 Misakachain/Qwen3.5-2B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Misakachain/Qwen3.5-2B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Misakachain/Qwen3.5-2B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Misakachain/Qwen3.5-2B:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Misakachain/Qwen3.5-2B:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Misakachain/Qwen3.5-2B:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Misakachain/Qwen3.5-2B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Misakachain/Qwen3.5-2B:Q4_K_M

Use Docker

docker model run hf.co/Misakachain/Qwen3.5-2B:Q4_K_M

LM Studio
Jan
Ollama
How to use Misakachain/Qwen3.5-2B with Ollama:
```
ollama run hf.co/Misakachain/Qwen3.5-2B:Q4_K_M
```
Unsloth Desktop

How to use Misakachain/Qwen3.5-2B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Misakachain/Qwen3.5-2B:Q4_K_M

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": "Misakachain/Qwen3.5-2B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Misakachain/Qwen3.5-2B with Docker Model Runner:
```
docker model run hf.co/Misakachain/Qwen3.5-2B:Q4_K_M
```

Lemonade

How to use Misakachain/Qwen3.5-2B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Misakachain/Qwen3.5-2B:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.5-2B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Misakachain/Qwen3.5-2B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Misakachain/Qwen3.5-2B:Q4_K_M

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 Misakachain/Qwen3.5-2B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Misakachain/Qwen3.5-2B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Misakachain/Qwen3.5-2B:Q4_K_M

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 "Misakachain/Qwen3.5-2B:Q4_K_M" \
  --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 — the model MISAKA testnet-11 mines with

This repository mirrors one exact file: the Q4_K_M GGUF that the MISAKA PALW consensus class CAT-M-0001 is pinned to. A PALW producer does not "run a Qwen model" — it runs this file, under this llama.cpp build, and the chain checks that it did. Any other quantization, any re-conversion, any newer upload of the same weights is a different artifact and will not produce a valid block.

That is why this mirror exists. The upstream weights are at Qwen/Qwen3.5-2B; what a miner needs is the byte-exact quantized file below.

The pin


File	`Qwen3.5-2B-Q4_K_M.gguf`
Size	`1280835840` bytes
SHA-256	`aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223`
Base model	`Qwen/Qwen3.5-2B` at revision `15852e8c16360a2fea060d615a32b45270f8a8fc`
llama.cpp	commit `030ebb558a5820b444a8f836ed5cdd46c9b4bd7a` (build 10358, unpatched)

Verify before you use it:

shasum -a 256 Qwen3.5-2B-Q4_K_M.gguf
# must print aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223

The worker checks this itself and refuses to load anything else, so a mismatch costs you a startup error rather than a stream of rejected blocks. Check it anyway: a file that fails here was truncated or replaced in transit, and knowing that is cheaper than debugging the worker.

Where it is registered

On testnet-11 this model is the consensus class

682756bc275c28cd8c2138cd62fb4be186e35e3c5cb97166f093bb9adb2799f6bc1a2749822390cfb9ff6d02a2dd65916fb208a05a00f34aa5819024940cffd4

which is derived from the model's own geometry — 24 layers, hidden 2048, 16 attention heads over 2 KV heads, vocabulary 248320, context 4096. Two models differing in any of those derive different class ids, which is what stops one being mined under the other's registration.

The runtime is part of the identity

A class is this file AND the binary that runs it. The chain pins a runtime_manifest_hash that covers the llama.cpp build flags, the floating-point environment, the thread count and this GGUF's digest together, and a producer whose worker reports a different one is refused before it can commit to anything.

Two consequences worth knowing before you build:

CPU architecture splits the class. x86_64 and aarch64 are registered separately (misaka-palw-lite-cpu/x86_64/v1 and misaka-palw-lite-cpu/aarch64-dotprod/v1) because the same weights on different SIMD paths produce different traces. A mixed fleet runs two classes and audits within each; it is not a bug, it is what makes an audit mean something.
A GPU build is a different class again. The CPU profile links no GPU or BLAS backend at all, and its identity says so.

Build the worker against the pinned llama.cpp commit above, point it at this file, and it will tell you which runtime class it is:

misaka-palw-worker --mode v2-manifest

Licence

Apache-2.0, inherited from Qwen/Qwen3.5-2B. This repository redistributes a quantization of those weights unmodified in substance; the upstream model card and licence terms apply.

Downloads last month: 16

GGUF

Model size

2B params

Architecture

qwen35

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Misakachain/Qwen3.5-2B

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B

Quantized

(182)

this model