Instructions to use zaindanaharper/flywheel-local-coder-14b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use zaindanaharper/flywheel-local-coder-14b 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 zaindanaharper/flywheel-local-coder-14b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf zaindanaharper/flywheel-local-coder-14b:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf zaindanaharper/flywheel-local-coder-14b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf zaindanaharper/flywheel-local-coder-14b:Q4_K_M

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 zaindanaharper/flywheel-local-coder-14b:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf zaindanaharper/flywheel-local-coder-14b:Q4_K_M

Use Docker

docker model run hf.co/zaindanaharper/flywheel-local-coder-14b:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use zaindanaharper/flywheel-local-coder-14b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zaindanaharper/flywheel-local-coder-14b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zaindanaharper/flywheel-local-coder-14b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/zaindanaharper/flywheel-local-coder-14b:Q4_K_M

Ollama
How to use zaindanaharper/flywheel-local-coder-14b with Ollama:
```
ollama run hf.co/zaindanaharper/flywheel-local-coder-14b:Q4_K_M
```
Unsloth Desktop

How to use zaindanaharper/flywheel-local-coder-14b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zaindanaharper/flywheel-local-coder-14b: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": "zaindanaharper/flywheel-local-coder-14b:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use zaindanaharper/flywheel-local-coder-14b with Docker Model Runner:
```
docker model run hf.co/zaindanaharper/flywheel-local-coder-14b:Q4_K_M
```

Lemonade

How to use zaindanaharper/flywheel-local-coder-14b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull zaindanaharper/flywheel-local-coder-14b:Q4_K_M

Run and chat with the model

lemonade run user.flywheel-local-coder-14b-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use zaindanaharper/flywheel-local-coder-14b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zaindanaharper/flywheel-local-coder-14b: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 zaindanaharper/flywheel-local-coder-14b:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use zaindanaharper/flywheel-local-coder-14b with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zaindanaharper/flywheel-local-coder-14b: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 "zaindanaharper/flywheel-local-coder-14b: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"

flywheel-local-coder-14b / MODEL_CARD.md

papacr0w

Refresh MODEL_CARD to shipped register (drop draft/do-not-publish + local paths)

148bd15 verified 4 months ago

preview code

Raw History Blame Contribute Delete

2.74 kB

Flywheel-Local-Coder-14B Model Card

A verified, retraceable continued-pretraining build on Qwen2.5-Coder-14B-Instruct. Identity and provenance are complete and re-checkable; benchmark evidence is attached, and no capability uplift over the base model is claimed.

Model identity

Release name: Flywheel-Local-Coder-14B
Model file: telos-coder-14b-cpt2020-q4_k_m.gguf
Base model: Qwen2.5-Coder-14B-Instruct (Alibaba Cloud / Qwen team, Apache-2.0)
Adapter: checkpoint-2020, QLoRA continued pretraining, final train loss 0.035
Composition: base weights merged with the adapter, then quantized
Quantization: Q4_K_M (GGUF)
Size: 8.99 GB (8,988,110,880 bytes)
Model SHA-256: 613db240e3efc6730f24042a4602d1f12f1c6b397af1d5a4d74f4e064d4064be
Adapter (safetensors) SHA-256: 4de07c6ea342d1cc200d4a6e2b28a63f6ee37f34c5c0926c35d8c7db74d38d0f
LoRA GGUF SHA-256: c89091709d7f385226000091dca976b7ce68086255e78af96599d06b6b52f547
Local serving name: Ollama flywheel-local-coder-14b

The full chain is recorded in provenance.json and tied to the downloaded bytes by checksums.sha256.

Training data

Continued pretraining on a 66.2-million-token corpus of 17,997 files from a real, working development ecosystem: production code, tests, documentation, and research notes. Corpus content hash 68345cdc6667f20d1678ac0a9139edc170348dfdebb9ae6045cde3d204f4fe62; pack shards hash 018798dfce7d4c86f5a6ea502a383553220f2e76facfe76acbe52b1c278ae543. Corpus source identifiers stay proprietary.

Intended use

Local-first coding: completion, small functions, refactors, test writing, and tool-calling on your own hardware, served via Ollama or llama.cpp. It pairs naturally with a propose-then-verify loop, which is how it is benchmarked.

Limitations

Q4_K_M quantization; quantization loss relative to the merged fp16 weights is not measured.
No capability uplift over the base model is claimed; our own measurement of that difference includes zero. See BENCHMARKS.md.
Refusal behavior, bias, and content boundaries are inherited from the base model and are not separately audited here.

License

Apache-2.0 derivative. Base model Qwen2.5-Coder-14B-Instruct is Copyright Alibaba Cloud, Apache-2.0. This artifact merges a locally trained QLoRA adapter into those weights and retains the Apache-2.0 license with attribution.

Benchmark status

Evidence attached. On the internal evaluation sets the model passes 8 of 8 baseline tasks and 8 of 10 deliberately contract-heavy hard tasks in a single attempt, each number with a confidence interval and the JSON to re-run it. See BENCHMARKS.md and the benchmarks/ folder.