Instructions to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF",
	filename="MTP/gemma-4-12B-it-MTP-BF16.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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M

Use Docker

docker model run hf.co/yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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": "yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M

Ollama
How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with Ollama:
```
ollama run hf.co/yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M
```

Unsloth Studio

How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF to start chatting

How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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": "yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-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 yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new
Docker Model Runner
How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with Docker Model Runner:
```
docker model run hf.co/yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M
```

Lemonade

How to use yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF-Q4_K_M

List all available models

lemonade list

A way to correctly reproduce a benchmark?

#40

by anonymousmaharaj - opened 7 days ago

Discussion

anonymousmaharaj

7 days ago

Hello dear friend! Thank you very much for this work!

I have been struggling for two days now and cannot correctly reproduce the benchmark specified in the model card (tau2-bench telecom). I just can't get it to work.
Could you tell me how you ran it and with what parameters? I want to run more queries. I think the community would be interested in this.

anonymousmaharaj changed discussion title from Способ корректно воспроизвести Benchmark to A way to correctly reproduce a benchmark? 7 days ago

yuxinlu1

Owner 6 days ago

Happy to help, and thanks! Two days of fighting this is almost always one of two small things: the --jinja flag, or
the served model name not matching what tau2 calls. Here's exactly how I ran it.

Serve the model with llama.cpp (the --jinja flag is essential — without it the Gemma 4 tool-calls come back as raw
text and every task fails; -a gemma sets the served name so tau2 can find it):

llama-server -m gemma4-v2-Q8_0.gguf -a gemma --host 127.0.0.1 --port 8000 -c 16384 -ngl 99 -fa on --jinja --temp 0.0

Point tau2 at that endpoint (tau2 uses litellm internally; the openai/ prefix + these env vars send it to your
local server):

OPENAI_API_BASE=http://127.0.0.1:8000/v1
OPENAI_API_KEY=dummy

Run the telecom domain (temp 0, 1 trial — at temp 0 extra trials are identical, so 1 is enough):

tau2 run --domain telecom --agent-llm openai/gemma --user-llm openai/gemma --num-trials 1 --num-tasks 20
--max-concurrency 1 --save-to tau2_telecom_v2

The gotchas that cost people days:

--jinja is mandatory. Before trusting any score, open one trace and confirm there are real tool_calls. If you see
tool calls printed as plain text, your template/--jinja is the problem — fix that first.
The gemma in -a gemma (server) must match openai/gemma (tau2 command), and OPENAI_API_BASE must point at the server.
If you re-run with the same --save-to, tau2 prompts (y/n) resume?; in a non-interactive shell that crashes with
EOFError. Delete the save dir first, or use a fresh --save-to name each run.
model=gemma isn't mapped yet from litellm is just a cost-estimation warning — harmless, ignore it.
Don't use --task-set-name telecom_small — it has a task-split bug; just use --domain telecom --num-tasks N.

Two honest caveats about the number (important if you want to publish):

I used the model as its own user-simulator (--user-llm openai/gemma). The user-sim model affects the score, so for a
clean, comparable benchmark across models, fix --user-llm to one strong model (a cheap cloud API works well) instead
of self-sim.
My published telecom run was effectively N=14 — I hit a handful of infrastructure/API errors that dropped some
tasks. So the robust signal is the relative pattern (v2 clearly beats base Gemma-4-12B on telecom), more than the
exact percentage.

To run more queries like you wanted: raise --num-tasks (telecom has more than 20 tasks), and if you want sampling
diversity rather than a single deterministic pass, raise the temperature and --num-trials together. Would love to see
what numbers you get — please share them in the thread.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment