Instructions to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("ajayk007/Qwen2.5-Coder-1.5B-Shellsmith")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

llama-cpp-python

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="ajayk007/Qwen2.5-Coder-1.5B-Shellsmith",
	filename="shellsmith-1.5b-f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith 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 ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16
# Run inference directly in the terminal:
llama cli -hf ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16
# Run inference directly in the terminal:
llama cli -hf ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

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 ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16
# Run inference directly in the terminal:
./llama-cli -hf ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

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 ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

Use Docker

docker model run hf.co/ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

LM Studio
Jan

vLLM

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

Ollama
How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with Ollama:
```
ollama run hf.co/ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16
```

Unsloth Studio

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith 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 ajayk007/Qwen2.5-Coder-1.5B-Shellsmith 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 ajayk007/Qwen2.5-Coder-1.5B-Shellsmith to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for ajayk007/Qwen2.5-Coder-1.5B-Shellsmith to start chatting

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"

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 ajayk007/Qwen2.5-Coder-1.5B-Shellsmith

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"

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 "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

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

MLX LM

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "ajayk007/Qwen2.5-Coder-1.5B-Shellsmith",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with Docker Model Runner:
```
docker model run hf.co/ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16
```

Lemonade

How to use ajayk007/Qwen2.5-Coder-1.5B-Shellsmith with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ajayk007/Qwen2.5-Coder-1.5B-Shellsmith:F16

Run and chat with the model

lemonade run user.Qwen2.5-Coder-1.5B-Shellsmith-F16

List all available models

lemonade list

Qwen2.5-Coder-1.5B-Shellsmith / README.md

ajayk007

Upload README.md with huggingface_hub

a24e6a9 verified 17 days ago

preview code

Raw

History Blame Contribute Delete

3.94 kB

	---
	license: apache-2.0
	base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct
	tags:
	- mlx
	- lora
	- text-generation
	- shell
	- command-line
	- code
	language:
	- en
	pipeline_tag: text-generation
	library_name: mlx
	---

	# Qwen2.5-Coder-1.5B-Shellsmith

	A small, fast model that turns **plain-English instructions into a single shell
	command** for macOS/Linux. LoRA fine-tune of
	[`Qwen/Qwen2.5-Coder-1.5B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct),
	trained and quantized end-to-end on an Apple Silicon Mac with
	[MLX](https://github.com/ml-explore/mlx).

	> "list files by size, biggest first" → `ls -lS`
	> "find files larger than 100 megabytes" → `find . -type f -size +100M`
	> "create a gzip tar archive of src named src.tar.gz" → `tar -czf src.tar.gz src`

	## Results

	Evaluated on a held-out test split the model never saw during training.
	Metrics are structural (no command execution) and conservative — see the
	[eval rubric](https://github.com/saiajay1/shellsmith/blob/main/eval/rubric.md).

	\| Model \| exact-match \| command-match \| flag-F1 \|
	\| --- \| :---: \| :---: \| :---: \|
	\| Base (Qwen2.5-Coder-1.5B-Instruct) \| 71% \| 83.9% \| 89.2% \|
	\| This model (LoRA) \| 100% \| 100% \| 100% \|

	command-match = correct program and* option-flag F1 ≥ 0.8.*

	> What the 100% means (and doesn't): the test split holds out unseen
	> phrasings, but the underlying task distribution (84 canonical tasks) overlaps
	> with training. So this measures reliable **in-distribution generalization across
	> wording** — the model consistently emits the canonical, idiomatic command
	> (`git add -A`, `ls -lS`, `git log --oneline -5`) where the base model drifts to
	> looser variants (`git add .`, `ls -lh \| sort -rh`, `git log -5`). It is not
	> evidence of generalization to entirely novel tasks; broadening the task set is
	> the obvious next step.

	## Usage

	### MLX (Apple Silicon)

	```bash
	pip install mlx-lm
	mlx_lm.generate --model ajayk007/Qwen2.5-Coder-1.5B-Shellsmith \
	--prompt "compress the logs folder into logs.tar.gz"
	```

	```python
	from mlx_lm import load, generate
	model, tok = load("ajayk007/Qwen2.5-Coder-1.5B-Shellsmith")
	messages = [
	{"role": "system", "content": "You are a shell command generator for macOS/Linux. "
	"Given a task in plain English, reply with a single safe shell command. "
	"Output only the command on one line, no explanation, no markdown."},
	{"role": "user", "content": "find all python files modified today"},
	]
	prompt = tok.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
	print(generate(model, tok, prompt=prompt, max_tokens=64))
	```

	### GGUF (llama.cpp / Ollama / LM Studio)

	A `shellsmith-1.5b-f16.gguf` file is included in this repo for use with llama.cpp-based runtimes.

	## Prompt format

	Uses the Qwen chat template with the system prompt shown above. Keep the system
	prompt for best results.

	## Training

	- Method: LoRA (rank 16), 16 layers, 400 iterations, batch size 4, lr 1e-4
	- Hardware: Apple M5 Pro (48 GB), MLX
	- Data: [`ajayk007/shellsmith-commands`](https://huggingface.co/datasets/ajayk007/shellsmith-commands) —
	curated (instruction, command) pairs with paraphrase augmentation, 80/10/10 split.

	## Limitations & safety

	- Generates commands across common categories (files, find/grep, archives, git,
	processes, networking). Outside this scope it falls back to base behavior.
	- Always read a generated command before running it. It can produce
	destructive commands (`rm`, `kill`, `chmod`) if you ask for them. There is no
	sandbox or confirmation step.
	- Single-command only; it does not write multi-step scripts.

	## Related
	Part of a series of focused "English → developer DSL" fine-tunes:
	- [Qwen2.5-Coder-7B-Querysmith](https://huggingface.co/ajayk007/Qwen2.5-Coder-7B-Querysmith) — schema-grounded text-to-SQL.

	## License

	Apache-2.0, inheriting from the base model.