Instructions to use Quobi/Quill with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Quobi/Quill with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Quobi/Quill",
	filename="quill-0.8b-Q4_K_M.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 Quobi/Quill 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 Quobi/Quill:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Quobi/Quill:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Quobi/Quill:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Quobi/Quill: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 Quobi/Quill:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Quobi/Quill: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 Quobi/Quill:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Quobi/Quill:Q4_K_M

Use Docker

docker model run hf.co/Quobi/Quill:Q4_K_M

LM Studio
Jan

vLLM

How to use Quobi/Quill with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Quobi/Quill:Q4_K_M

Ollama
How to use Quobi/Quill with Ollama:
```
ollama run hf.co/Quobi/Quill:Q4_K_M
```

Unsloth Studio

How to use Quobi/Quill 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 Quobi/Quill 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 Quobi/Quill to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Quobi/Quill to start chatting

How to use Quobi/Quill with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Quobi/Quill:Q4_K_M

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": "Quobi/Quill:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Quobi/Quill with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Quobi/Quill: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 Quobi/Quill:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use Quobi/Quill with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Quobi/Quill: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 "Quobi/Quill: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"

Docker Model Runner
How to use Quobi/Quill with Docker Model Runner:
```
docker model run hf.co/Quobi/Quill:Q4_K_M
```

Lemonade

How to use Quobi/Quill with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Quobi/Quill:Q4_K_M

Run and chat with the model

lemonade run user.Quill-Q4_K_M

List all available models

lemonade list

ZentraAI commited on Jun 8

Commit

4cc2cc3

verified ·

1 Parent(s): 03030ad

docs: remove em dashes from the model card

Browse files

Files changed (1) hide show

README.md +8 -8

README.md CHANGED Viewed

@@ -20,13 +20,13 @@ pipeline_tag: text-generation
 inference: false
 ---
-# Quill — on-device dictation cleanup models
 **Quill** is a family of small language models that turn raw speech-to-text
-output into clean, written text — **entirely on your own device**. It removes
 filler words (*um*, *uh*, *like*, *you know*), fixes punctuation and
 capitalization, repairs spoken self-corrections and false starts, and collapses
-the stutters and repeats that dictation produces — without changing your words
 or sending anything to the cloud.
 Quill is the cleanup stage of **[Quobi](https://huggingface.co/quobi)**, a
@@ -43,7 +43,7 @@ Quill rewrites that into what you actually meant to write:
 > **"So I was thinking maybe we could meet up at three."**
-It is **not** a chatbot and not an instruction-following assistant — it does one
 job: clean dictated text. Feeding it questions or commands will not get answers;
 it will just clean the text.
@@ -52,7 +52,7 @@ it will just clean the text.
 Quill is a fine-tune of **[Qwen3.5](https://huggingface.co/Qwen)** by the Qwen
 team (Alibaba), used under the **Apache 2.0** license. Qwen3.5 is a hybrid
 architecture interleaving **Mamba-2 / state-space (SSM)** layers with periodic
-full-attention layers, which makes the small sizes fast and memory-light —
 well suited to on-device, low-latency cleanup. All credit for the base models
 goes to the Qwen team; Quill only adds task-specific fine-tuning.
@@ -66,7 +66,7 @@ goes to the Qwen team; Quill only adds task-specific fine-tuning.
 | Tier | Best for | Behavior |
 |---|---|---|
-| **0.8B** | Phones and any CPU (recommended default) | **Verbatim** — faithful cleanup, no rephrasing |
 | **2B** | Mid-range machines / a modest GPU | Verbatim + light tidying |
 | **4B** | Desktops with a GPU | Verbatim + tidying + light formatting |
@@ -82,7 +82,7 @@ the larger tiers take on more rewriting and structure.
 llama-server -m quill-0.8b-Q4_K_M.gguf --host 127.0.0.1 --port 8080 -ngl 99
 ```
-**Prompt format — important.** Use ChatML with the assistant turn pre-seeded
 with an **empty think block** so the model does not emit chain-of-thought:
 ```
@@ -99,7 +99,7 @@ yeah so um the meeting is gonna be like at uh three thirty tomorrow i think<|im_
 → **"The meeting is at 3:30 tomorrow."**
-> ⚠️ Do **not** pass `--jinja` — it re-enables chain-of-thought leakage. Use the
 > raw prompt above (or the `/completion` endpoint) with the pre-seeded empty
 > `<think></think>` block. Greedy decoding (`temperature = 0`) is recommended.

 inference: false
 ---
+# Quill: on-device dictation cleanup models
 **Quill** is a family of small language models that turn raw speech-to-text
+output into clean, written text, **entirely on your own device**. It removes
 filler words (*um*, *uh*, *like*, *you know*), fixes punctuation and
 capitalization, repairs spoken self-corrections and false starts, and collapses
+the stutters and repeats that dictation produces, without changing your words
 or sending anything to the cloud.
 Quill is the cleanup stage of **[Quobi](https://huggingface.co/quobi)**, a
 > **"So I was thinking maybe we could meet up at three."**
+It is **not** a chatbot and not an instruction-following assistant. It does one
 job: clean dictated text. Feeding it questions or commands will not get answers;
 it will just clean the text.
 Quill is a fine-tune of **[Qwen3.5](https://huggingface.co/Qwen)** by the Qwen
 team (Alibaba), used under the **Apache 2.0** license. Qwen3.5 is a hybrid
 architecture interleaving **Mamba-2 / state-space (SSM)** layers with periodic
+full-attention layers, which makes the small sizes fast and memory-light,
 well suited to on-device, low-latency cleanup. All credit for the base models
 goes to the Qwen team; Quill only adds task-specific fine-tuning.
 | Tier | Best for | Behavior |
 |---|---|---|
+| **0.8B** | Phones and any CPU (recommended default) | **Verbatim**: faithful cleanup, no rephrasing |
 | **2B** | Mid-range machines / a modest GPU | Verbatim + light tidying |
 | **4B** | Desktops with a GPU | Verbatim + tidying + light formatting |
 llama-server -m quill-0.8b-Q4_K_M.gguf --host 127.0.0.1 --port 8080 -ngl 99
 ```
+**Prompt format (important).** Use ChatML with the assistant turn pre-seeded
 with an **empty think block** so the model does not emit chain-of-thought:
 ```
 → **"The meeting is at 3:30 tomorrow."**
+> ⚠️ Do **not** pass `--jinja`. It re-enables chain-of-thought leakage. Use the
 > raw prompt above (or the `/completion` endpoint) with the pre-seeded empty
 > `<think></think>` block. Greedy decoding (`temperature = 0`) is recommended.