Instructions to use MicroFlare/microFlare-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use MicroFlare/microFlare-v1 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 MicroFlare/microFlare-v1
# Run inference directly in the terminal:
llama cli -hf MicroFlare/microFlare-v1

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf MicroFlare/microFlare-v1
# Run inference directly in the terminal:
llama cli -hf MicroFlare/microFlare-v1

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 MicroFlare/microFlare-v1
# Run inference directly in the terminal:
./llama-cli -hf MicroFlare/microFlare-v1

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 MicroFlare/microFlare-v1
# Run inference directly in the terminal:
./build/bin/llama-cli -hf MicroFlare/microFlare-v1

Use Docker

docker model run hf.co/MicroFlare/microFlare-v1

LM Studio
Jan
Ollama
How to use MicroFlare/microFlare-v1 with Ollama:
```
ollama run hf.co/MicroFlare/microFlare-v1
```

Unsloth Studio

How to use MicroFlare/microFlare-v1 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for MicroFlare/microFlare-v1 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for MicroFlare/microFlare-v1 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for MicroFlare/microFlare-v1 to start chatting

How to use MicroFlare/microFlare-v1 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MicroFlare/microFlare-v1

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "MicroFlare/microFlare-v1"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

OpenClaw new

How to use MicroFlare/microFlare-v1 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MicroFlare/microFlare-v1

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 "MicroFlare/microFlare-v1" \
  --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"

Docker Model Runner
How to use MicroFlare/microFlare-v1 with Docker Model Runner:
```
docker model run hf.co/MicroFlare/microFlare-v1
```

Lemonade

How to use MicroFlare/microFlare-v1 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull MicroFlare/microFlare-v1

Run and chat with the model

lemonade run user.microFlare-v1-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use MicroFlare/microFlare-v1 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MicroFlare/microFlare-v1

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 MicroFlare/microFlare-v1

Run Hermes

hermes

Atomic Chat

microFlare v1

microFlare v1 is an asymmetrically quantized version of Qwen 3.6 35B-A3B. The model has been optimized layer by layer, with high precision for important parts and lower bitrates for less consequential weights. It is designed so it can be ran on low spec hardware, as low as a 4GB GPU + 16GB of system RAM utilizing CPU offloading of experts. It should hopefully fit well onto a 16GB GPU without need for CPU offload, but this has not yet been tested. Additionally if you don't even have a 4GB GPU, it can be ran completely on CPU with 16GB of system RAM, but it will be slow probably.

Note that the MTP (layer 64) predictor has been stripped out of this version, as it proved to be detrimental on older hardware. I am considering releasing another version with the MTP included, or maybe just a stand alone file for it (let me know which would be more useful to you).

Perplexity drift

Ran with a 2k context, revision A has a relative perplexity score* of 6.2761 +/- 0.03961
*note scoring on my machine is different than as reported elsewhere, so this is better than it may sound at first

Model	Size	Perplexity Score
bartowski/Q4_K	22GB	5.9654 +/- 0.03726
unsloth/UD-IQ3_S	13.7GB	6.2496 +/- 0.03948
microFlare v1.a	14GB	6.2761 +/- 0.03961
mudler/APEX-I-Mini	14.3GB	6.3554 +/- 0.04057
bartowski/Q2_K_L	14GB	6.4180 +/- 0.04066

Test Suites

GSM8K (basic math)

Ran with 5 shot and 2048 context

Match type	Score
flexible	0.5906 ±0.0135
strict	0.6058 ±0.0135

MMLU Pro (comprehensive)*

*note these results are from limited partial run, only the first 20 questions of the 14 categories were tested

Ran with 5 shot and 8196 context

Tasks	Version	Filter	n-shot	Metric	Value	Stderr
Total Average	2.0	custom-extract	5	exact_match	0.7286	±0.0262
- biology	3.1	custom-extract	5	exact_match	0.8000	±0.0918
- business	3.1	custom-extract	5	exact_match	0.7500	±0.0993
- chemistry	3.1	custom-extract	5	exact_match	0.8500	±0.0819
- computer_science	3.1	custom-extract	5	exact_match	0.8500	±0.0819
- economics	3.1	custom-extract	5	exact_match	0.6500	±0.1094
- engineering	3.1	custom-extract	5	exact_match	0.5500	±0.1141
- health	3.1	custom-extract	5	exact_match	0.7500	±0.0993
- history	3.1	custom-extract	5	exact_match	0.7000	±0.1051
- law	3.1	custom-extract	5	exact_match	0.4500	±0.1141
- math	3.1	custom-extract	5	exact_match	0.9000	±0.0688
- other	3.1	custom-extract	5	exact_match	0.6000	±0.1124
- philosophy	3.1	custom-extract	5	exact_match	0.8000	±0.0918
- physics	3.1	custom-extract	5	exact_match	0.7000	±0.1051
- psychology	3.1	custom-extract	5	exact_match	0.8500	±0.0819

Manual prompt testing

So far all testing has been successful. Of the 50 some odd prompts it's been tested with everyone came back with a high quality answer/result. It built a pretty decent test website on a one shot, so coding abilities are intact. It has not been tested with any tools or agent harnesses yet, nor image based inputs. I've only tested text so far.

For more details and tips on running optimally, see my full blog post.

Downloads last month: 68

GGUF

Model size

35B params

Architecture

qwen35moe

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

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

Model tree for MicroFlare/microFlare-v1

Base model

Qwen/Qwen3.6-35B-A3B

Quantized

(745)

this model