Instructions to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual 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 lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual: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 lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual: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 lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M

Use Docker

docker model run hf.co/lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M

LM Studio
Jan

vLLM

How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M

Ollama
How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with Ollama:
```
ollama run hf.co/lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M
```
Unsloth Desktop

How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual: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": "lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with Docker Model Runner:
```
docker model run hf.co/lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M
```

Lemonade

How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M

Run and chat with the model

lemonade run user.Frame2KG-Qwen3.5-2b-CT_Textual-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual: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 lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual: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 "lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual: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"

Frame2KG-Qwen3.5-2b-CT_Textual

This repository contains GGUF quantised files for the Frame2KG fine-tuned Qwen3.5 2b model.

These GGUF releases are optimized deployment variants of Frame2KG models. They are provided for practical inference use and may not exactly match the original weights, checkpoints, or evaluation configuration reported in the Frame2KG paper.

Model Details

Family: Qwen3.5
Size: 2b
Output format: textual compressed Frame2KG graph strings (literal CT delimiters, no added tokenizer vocabulary)
Base model: Qwen/Qwen3.5-2B
Model type: qwen3_5
Architecture: Qwen3_5ForConditionalGeneration
Fine-tuning method: PEFT LoRA
LoRA rank: 8
LoRA alpha: 16
Tokenizer extension: none; CT delimiters are ordinary text

Files

File	Size	Notes
`Frame2KG-Qwen3.5-2b-CT_Textual.f16.gguf`	3.52 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.IQ4_XS.gguf`	1.12 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q2_K.gguf`	923.67 MB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q3_K_L.gguf`	1.08 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q3_K_M.gguf`	1.02 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q3_K_S.gguf`	972.91 MB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q4_K_M.gguf`	1.19 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q4_K_S.gguf`	1.13 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q5_K_M.gguf`	1.31 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q5_K_S.gguf`	1.28 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q6_K.gguf`	1.45 GB	model weights
`Frame2KG-Qwen3.5-2b-CT_Textual.Q8_0.gguf`	1.87 GB	model weights
`mmproj-Frame2KG-Qwen3.5-2b-CT_Textual.f16.gguf`	637.27 MB	multimodal projector
`mmproj-Frame2KG-Qwen3.5-2b-CT_Textual.Q8_0.gguf`	347.77 MB	multimodal projector

CT_Textual Metadata

This CT_Textual variant uses the CT output structure and CT training data, but the graph delimiters are handled as ordinary text by the base tokenizer. No learned single-token CT marker vocabulary is added.

Literal delimiters:

<|graph|>
<|nodes|>
<|node|>
<|id|>
<|label|>
<|bbox|>
<|conf|>
<|attrs|>
<|edges|>
<|edge|>
<|src|>
<|pred|>
<|tgt|>
<|end_graph|>

Usage Notes

These files are intended for llama.cpp-compatible GGUF runtimes.

Scope

Frame2KG models are intended to convert image or frame content into structured graph-style outputs. The exact output format depends on the variant:

JSON variants target JSON-formatted Frame2KG output.
CT variants target compressed Frame2KG graph tokens.
CT_Textual variants target the CT graph structure with literal marker text and no added tokenizer vocabulary.

Citation

If you use this model in your work, please cite the paper:

@inproceedings{watson2026frame2kg,
  title = {Frame2KG: A Benchmark and Evaluation Toolkit for Interpretable Frame-to-Graph Generation},
  author = {Watson, Lewis N. and Strathearn, Carl and Mitchell, Kenny and Yu, Yanchao},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)},
  month = {May},
  year = {2026},
  pages = {10912--10926},
  address = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy and Krek, Simon and Toral, Antonio},
  doi = {10.63317/4ys6kofrzoc5},
  url = {https://doi.org/10.63317/4ys6kofrzoc5}
}

Disclaimer

This model is provided as is, without warranties or guarantees of any kind, either express or implied. The authors make no representations regarding the accuracy, reliability, safety, suitability, or performance of the model or its outputs.

The model may generate incorrect, incomplete, or misleading results and should not be relied upon for critical, safety-sensitive, legal, medical, financial, or other high-stakes decisions. Use of this model is entirely at your own risk.

The authors accept no liability for damages, losses, or consequences arising from use, misuse, or inability to use the model.

Downloads last month: 78

GGUF

Model size

2B params

Architecture

qwen35

Hardware compatibility

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Model tree for lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B

Quantized

(240)

this model

lewiswatson
/

Frame2KG-Qwen3.5-2b-CT_Textual

Frame2KG-Qwen3.5-2b-CT_Textual

Model Details

Files

CT_Textual Metadata

Usage Notes

Scope

Citation

Disclaimer

Model tree for lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual

Datasets used to train lewiswatson/Frame2KG-Qwen3.5-2b-CT_Textual