Instructions to use LaraAI-Labs/tellama with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use LaraAI-Labs/tellama 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 LaraAI-Labs/tellama:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf LaraAI-Labs/tellama:Q4_K_M

Install from WinGet (Windows)

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

Use Docker

docker model run hf.co/LaraAI-Labs/tellama:Q4_K_M

LM Studio
Jan

vLLM

How to use LaraAI-Labs/tellama with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/LaraAI-Labs/tellama:Q4_K_M

Ollama
How to use LaraAI-Labs/tellama with Ollama:
```
ollama run hf.co/LaraAI-Labs/tellama:Q4_K_M
```
Unsloth Desktop

How to use LaraAI-Labs/tellama with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LaraAI-Labs/tellama: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": "LaraAI-Labs/tellama:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use LaraAI-Labs/tellama with Docker Model Runner:
```
docker model run hf.co/LaraAI-Labs/tellama:Q4_K_M
```

Lemonade

How to use LaraAI-Labs/tellama with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull LaraAI-Labs/tellama:Q4_K_M

Run and chat with the model

lemonade run user.tellama-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use LaraAI-Labs/tellama with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LaraAI-Labs/tellama: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 LaraAI-Labs/tellama:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use LaraAI-Labs/tellama with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf LaraAI-Labs/tellama: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 "LaraAI-Labs/tellama: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"

Tellama model distributions

The distribution below passed Tellama's stated model qualification gates. App release and update-installation acceptance are tracked separately.

TM Qwen3 4B Q4_K_M — Tellama quantization

Tellama quantized the original Qwen3 4B weights through F16 into Q4_K_M, retaining Q8_0 token embeddings and output tensors. No fine-tuning or calibration was used. Qwen/Alibaba Cloud retains authorship of the base model. Tellama's validation does not imply Qwen endorsement or guarantee correct answers.

Exact file: models/qwen3-4b-tellama-thinking-q4_k_m.gguf
Bytes: 2,591,481,472 (approximately 2.59 GB download).
SHA-256: 01e739de8a200c769e72a676d038b84b1c042bd160a153a348fbb443bd713b85.
Base revision: 1cfa9a7208912126459214e8b04321603b3df60c.
llama.cpp revision: cb295bf59663cd3577389315636772f4060bd1f5.
License and attribution: licenses/qwen3-4b-tellama-thinking-q4_k_m/.
Reproducible recipe: recipes/qwen3-4b-tellama-thinking-q4_k_m.json.

Measured support and limitations

Measured on Samsung Galaxy Z Fold6 SM-F956N / Android 16, CPU, Tellama DEEP mode, 4,096-token context, F16 KV cache, default 768-token output budget, temperature 0.7, top-p 0.9 and top-k 40. Explicit user profiles can differ from these settings.

Screen	Result
Basic, KV reset each turn	21/21
Basic, reuse allowed	20/21
Extended, multi-turn and long context	13/13

All critical checks passed; 54 of 55 total checks passed. The noncritical failure translated the requested Korean JSON city value into English (Busan). Do not interpret this small synthetic suite as universal accuracy or structured-output certification. Review important answers independently.

Minimum measured decode speed was 3.76 tokens/s. Visible responses sometimes took over two minutes because of internal reasoning. Basic short prompts were fully re-prefilled (no actual KV reuse); the extended suite reused up to 1,122 tokens. Peak observed Android thermal status was 2. Memory/thermal guards remain enabled; these measurements do not guarantee sustained performance under every condition.

The original quality reports truthfully identify an internal modified 1.3.8 benchmark build. They are not reports from the previously published stock 1.3.8. Internal 1.3.9 hosting acceptance completed a full download, interruption/resume, process restart, SHA-256 verification and installation on Android16/16KB emulator. Fold6 1.3.9 runtime replacement from a loaded Qwen2.5 1.5B to this 4B and back passed at 4,096 context, without changing user settings. These are distinct from the final signed APK/update release checks.

FAST mode, GPU execution, other devices, image/audio input, and the original upstream maximum context are not qualified by these results. Tellama requires explicit consent to enable DEEP for this distribution. An ordinary Qwen3 base model can technically run without thinking; this curated variant is offered only under the settings that passed Tellama's checks.

Previous experimental files

The previous Qwen3.5 4B file remains available for provenance/history, but its extended arithmetic checks failed. It is not a TM-approved recommendation and is not offered for new downloads in the new curated catalog. Existing users' local models and conversations are not automatically deleted.

Downloads last month: 118

GGUF

Model size

4B params

Architecture

qwen3

Hardware compatibility

4-bit

Model tree for LaraAI-Labs/tellama

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Quantized

(335)

this model