Instructions to use hudsongouge/tmax-2b-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use hudsongouge/tmax-2b-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="hudsongouge/tmax-2b-GGUF",
	filename="tmax-2b-Q2_K.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 hudsongouge/tmax-2b-GGUF 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 hudsongouge/tmax-2b-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf hudsongouge/tmax-2b-GGUF:Q4_K_M

Install from WinGet (Windows)

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

Use Docker

docker model run hf.co/hudsongouge/tmax-2b-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use hudsongouge/tmax-2b-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hudsongouge/tmax-2b-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hudsongouge/tmax-2b-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/hudsongouge/tmax-2b-GGUF:Q4_K_M

Ollama
How to use hudsongouge/tmax-2b-GGUF with Ollama:
```
ollama run hf.co/hudsongouge/tmax-2b-GGUF:Q4_K_M
```

Unsloth Studio

How to use hudsongouge/tmax-2b-GGUF 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 hudsongouge/tmax-2b-GGUF 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 hudsongouge/tmax-2b-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for hudsongouge/tmax-2b-GGUF to start chatting

How to use hudsongouge/tmax-2b-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hudsongouge/tmax-2b-GGUF: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": "hudsongouge/tmax-2b-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use hudsongouge/tmax-2b-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hudsongouge/tmax-2b-GGUF: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 hudsongouge/tmax-2b-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new
Docker Model Runner
How to use hudsongouge/tmax-2b-GGUF with Docker Model Runner:
```
docker model run hf.co/hudsongouge/tmax-2b-GGUF:Q4_K_M
```

Lemonade

How to use hudsongouge/tmax-2b-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull hudsongouge/tmax-2b-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.tmax-2b-GGUF-Q4_K_M

List all available models

lemonade list

allenai/tmax-2b GGUF

GGUF quantizations of allenai/tmax-2b, a terminal-agent model trained with DPPO on top of Qwen 3.5 2B.

Important: MTP head excluded

The original allenai/tmax-2b checkpoint is a Qwen3_5ForConditionalGeneration model whose config.json declares mtp_num_hidden_layers: 1, but the shipped model.safetensors does not contain the MTP head weights. Converting with the default MTP settings produces a GGUF that fails to load with:

error loading model: missing tensor 'blk.24.attn_norm.weight'

These GGUFs were converted with --no-mtp so the nextn_predict_layers metadata is omitted and the model loads correctly in llama.cpp / LM Studio / any GGUF-compatible runtime that supports the qwen35 architecture. MTP speculative decoding is therefore not available for this checkpoint.

Quantizations

Name	Size	BPW	Notes
Q8_0	1.9 GB	8.5	Highest quality quantized
Q6_K	1.4 GB	6.6	Very high quality
Q5_K_M	1.3 GB	5.7	High quality, good balance
Q4_K_M	1.2 GB	5.4	Recommended default
Q4_K_S	1.1 GB	4.9	Smaller than Q4_K_M
Q3_K_M	1.0 GB	4.3	Lower quality, smaller
Q2_K	924 MB	2.6	Smallest, significant quality loss

For most users, Q4_K_M is the recommended balance of quality and size.

Conversion details

Source: allenai/tmax-2b (BF16 safetensors, 3.5 GB)
Tool: llama.cpp convert_hf_to_gguf.py (build 9820) with --no-mtp --outtype bf16
Quantization: llama-quantize from BF16 to each target type
Architecture: qwen35 (hybrid linear/full attention, mRoPE)

Runtime requirements

Requires a GGUF runtime with qwen35 architecture support:

llama.cpp build >= ~9300 (b9334+ confirmed working)
LM Studio 0.4.12+ with llama.cpp runtime >= 2.5.1
Any runtime advertising Qwen 3.5 support

Usage

# llama.cpp
llama-cli -m tmax-2b-Q4_K_M.gguf -p "Hello" -n 256 -ngl 99

# LM Studio (CLI)
lms import tmax-2b-Q4_K_M.gguf -y -c --user-repo local/tmax-2b
lms load tmax-2b
lms chat tmax-2b

License

Apache 2.0, inherited from the base model. See allenai/tmax-2b for the original model card and responsible-use guidelines.

Downloads last month: 126

GGUF

Model size

2B params

Architecture

qwen35

Hardware compatibility

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

Model tree for hudsongouge/tmax-2b-GGUF

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B

Finetuned

allenai/tmax-2b

Quantized

(1)

this model