### Quick Start Installation and Setup

Source: https://github.com/liquid4all/cookbook/blob/main/examples/localcowork/README.md

Steps to clone the repository, set up development environment, download the LFM2-24B-A2B model, and start the model server and application.

```bash
# 1. Clone and set up
git clone <repo-url> && cd localCoWork
./scripts/setup-dev.sh

# 2. Download LFM2-24B-A2B (~14 GB)
#    https://huggingface.co/LiquidAI/LFM2-24B-A2B-GGUF
pip install huggingface-hub
python3 -c "
from huggingface_hub import hf_hub_download
hf_hub_download('LiquidAI/LFM2-24B-A2B-GGUF',
                'LFM2-24B-A2B-Q4_K_M.gguf',
                local_dir='$HOME/Projects/_models/')
"

# 3. Start the model server
./scripts/start-model.sh

# 4. Launch the app (in another terminal)
cargo tauri dev

```

--------------------------------

### Quick Start Development Setup

Source: https://github.com/liquid4all/cookbook/blob/main/examples/localcowork/CONTRIBUTING.md

Steps to clone the repository, set up the development environment, verify the setup, and start the application in development mode.

```bash
# 1. Fork and clone
git clone <your-fork-url> && cd localCoWork

# 2. Run setup (installs all deps, creates venvs, checks prereqs)
./scripts/setup-dev.sh

# 3. Verify everything works
./scripts/smoke-test.sh

# 4. Start the app in dev mode
cargo tauri dev
```

--------------------------------

### Install and Authenticate Modal CLI

Source: https://github.com/liquid4all/cookbook/blob/main/examples/voice-chat/README.md

Installs the Modal CLI using uv and then authenticates the user with the Modal service.

```bash
uv add modal
uv run modal token new
```

--------------------------------

### Inference Setup

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/sft_for_vision_language_model.ipynb

Sets up the model for inference and prepares inputs for the first example.

```python
FastVisionModel.for_inference(model) # Enable for inference!

image = dataset[2]["image"]
instruction = "Write the LaTeX representation for this image."

messages = [
    {"role": "user", "content": [
        {"type": "image"},
        {"type": "text", "text": instruction}
    ]}
]
input_text = tokenizer.apply_chat_template(messages, add_generation_prompt = True)
inputs = tokenizer(
    image,
    input_text,
    add_special_tokens = False,
    return_tensors = "pt",
).to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer, skip_prompt = True)
_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 128,
                   use_cache = True, temperature = 1.5, min_p = 0.1)
```

--------------------------------

### Setup Dependencies

Source: https://github.com/liquid4all/cookbook/blob/main/examples/local-coding-assistant/README.md

Clones the project, installs dependencies using uv, and sets up the environment file with an API key.

```bash
cd examples/local-coding-assistant

uv sync

cp .env.example .env
# Edit .env and set ANTHROPIC_API_KEY
```

--------------------------------

### Initial Setup and Installation

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/grpo_with_unsloth.ipynb

Installs necessary libraries like unsloth and vllm, and sets an environment variable for Unsloth's VLLM standby mode.

```python
%%capture
import os
os.environ["UNSLOTH_VLLM_STANDBY"] = "1" # [NEW] Extra 30% context lengths!
if "COLAB_" not in "".join(os.environ.keys()):
    # If you're not in Colab, just use pip install or uv pip install
    !pip install unsloth vllm
else:
    pass # For Colab / Kaggle, we need extra instructions hidden below \/ 
```

--------------------------------

### Setup Development Environment

Source: https://github.com/liquid4all/cookbook/blob/main/examples/localcowork/docs/contributing.md

Clones the repository and runs the setup script to install dependencies and configure the development environment.

```bash
git clone <repo-url> && cd localCoWork
./scripts/setup-dev.sh
```

--------------------------------

### One-time setup

Source: https://github.com/liquid4all/cookbook/blob/main/examples/home-assistant/README.md

Clones the leap-finetune repository, installs dependencies, and authenticates with HuggingFace and Modal.

```bash
git clone https://github.com/Liquid4All/leap-finetune
cd leap-finetune && uv sync && cd -
huggingface-cli login
modal setup
```

--------------------------------

### Quick start: Clone repository and run locally

Source: https://github.com/liquid4all/cookbook/blob/main/examples/meeting-summarization/README.md

This command clones the repository, navigates to the example directory, and runs the summarization CLI with specific model and transcript file arguments.

```sh
git clone https://github.com/Liquid4All/cookbook.git
cd cookbook/examples/meeting-summarization
uv run summarize.py \
  --model LiquidAI/LFM2-2.6B-Transcript-GGUF \
  --hf-model-file LFM2-2.6B-Transcript-1-GGUF.gguf \
  --transcript-file transcripts/example_1.txt
```

--------------------------------

### Environment Variable Setup

Source: https://github.com/liquid4all/cookbook/blob/main/examples/voice-assistant/README.md

Instructions to copy the example environment file and add the HF_TOKEN.

```bash
cp .env.example .env
# then edit .env and replace hf_xxx with your token
```

--------------------------------

### Quick Start: Run Published Fine-tuned Model

Source: https://github.com/liquid4all/cookbook/blob/main/examples/voice-assistant/README.md

Commands to clone the repository, install dependencies, and run the evaluation script for the published fine-tuned model.

```bash
git clone https://github.com/Liquid4All/cookbook
cd cookbook/examples/voice-assistant
uv sync
uv run python scripts/eval.py --config configs/finetuned-q8.yaml
```

--------------------------------

### Run Locally

Source: https://github.com/liquid4all/cookbook/blob/main/examples/hand-voice-racer/README.md

Commands to install dependencies and start the development server for the Hand & Voice Racer project.

```bash
npm install
npm run dev
```

--------------------------------

### Install dependencies

Source: https://github.com/liquid4all/cookbook/blob/main/examples/audio-webgpu-demo/README.md

Installs project dependencies specified in package.json using npm.

```sh
npm install
```

--------------------------------

### One-time setup for fine-tuning

Source: https://github.com/liquid4all/cookbook/blob/main/examples/home-assistant/CLAUDE.md

Clones the fine-tuning repository, installs dependencies, logs into Hugging Face, and sets up Modal.

```bash
git clone https://github.com/Liquid4All/leap-finetune
cd leap-finetune && uv sync && cd -

huggingface-cli login
modal setup
```

--------------------------------

### Install Dependencies

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/cpt_text_completion_with_unsloth.ipynb

Installs necessary libraries for fine-tuning, with conditional installation for Google Colab.

```python
%%capture
import os, re
if "COLAB_" not in "".join(os.environ.keys()):
    !pip install unsloth
else:
    # Do this only in Colab notebooks! Otherwise use pip install unsloth
    import torch; v = re.match(r"[0-9]{1,}\.[0-9]{1,}", str(torch.__version__)).group(0)
    xformers = "xformers==" + ("0.0.33.post1" if v=="2.9" else "0.0.32.post2" if v=="2.8" else "0.0.29.post3")
    !pip install --no-deps bitsandbytes accelerate {xformers} peft trl triton cut_cross_entropy unsloth_zoo
    !pip install sentencepiece protobuf "datasets==4.3.0" "huggingface_hub>=0.34.0" hf_transfer
    !pip install --no-deps unsloth
!pip install transformers==4.57.3
!pip install --no-deps trl==0.22.2
```

--------------------------------

### Install uv

Source: https://github.com/liquid4all/cookbook/blob/main/examples/audio-transcription-cli/README.md

Installs the uv package manager on macOS/Linux or Windows.

```sh
curl -LsSf https://astral.sh/uv/install.sh | sh
```

```powershell
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
```

--------------------------------

### Print First Dataset Example

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/cpt_translation_with_unsloth.ipynb

Prints the first example from the loaded Alpaca dataset to show its structure.

```python
print(alpaca_dataset[0])
```

--------------------------------

### Quick start commands

Source: https://github.com/liquid4all/cookbook/blob/main/examples/audio-car-cockpit/README.md

Commands to set up the Python environment, prepare the audio and tool calling models, and launch the demo.

```bash
# Setup python env
make setup

# Optional: if llama-server is already in your PATH, symlink it instead of building
# ln -s $(which llama-server) llama-server
# Note: when building for ROCm, also install: sudo apt install -y libstdc++-14-dev

# Prepare the audio and tool calling models
make LFM2.5-Audio-1.5B-GGUF LFM2-1.2B-Tool-GGUF

# Launch demo
make -j2 audioserver serve
```

--------------------------------

### System Prompt Example

Source: https://github.com/liquid4all/cookbook/blob/main/examples/localcowork/docs/patterns/context-window-management.md

The system prompt is loaded once at conversation start and never evicted.

```text
You are LocalCowork, a private desktop assistant. You help users manage files,
documents, emails, calendars, and tasks — entirely on their local machine.

Rules:
1. Never write code. Use only the tools provided.
2. For mutable actions, show preview and wait for confirmation.
3. For destructive actions, show explicit warning.
4. Explain what you'll do before calling tools.
5. For multi-step tasks, plan first, then execute step by step.
```

--------------------------------

### Setup Before Webinar

Source: https://github.com/liquid4all/cookbook/blob/main/examples/local-coding-assistant/MEETING_INTELLIGENCE.md

Commands to set up the environment for the live demo, including syncing dependencies and copying the .env file.

```bash
cd examples/meeting-intelligence-agent
uv sync
cp .env.example .env
# Add ANTHROPIC_API_KEY to .env (for the Anthropic backend demo)

```

--------------------------------

### Start the development server

Source: https://github.com/liquid4all/cookbook/blob/main/examples/audio-webgpu-demo/README.md

Starts the development server, typically providing a local URL like http://localhost:5173 to access the app in the browser.

```sh
npm run dev
```

--------------------------------

### Install and authenticate Modal

Source: https://github.com/liquid4all/cookbook/blob/main/examples/satellite-vlm/README.md

Installs dependencies and logs into Modal and Hugging Face. This is the first step in setting up the environment for fine-tuning.

```bash
uv sync
uv run python -m modal setup
uv run huggingface-cli login
```

--------------------------------

### Install XcodeGen

Source: https://github.com/liquid4all/cookbook/blob/main/examples/telco-triage-ios/README.md

Installs XcodeGen using Homebrew if it's not already installed.

```bash
brew install xcodegen
```

--------------------------------

### Starting a Model Server

Source: https://github.com/liquid4all/cookbook/blob/main/examples/localcowork/docs/model-analysis/README.md

Commands to start a local model server using llama-server or Ollama, which are prerequisites for running benchmarks.

```bash
# LFM2-24B via llama-server (default port 8080)
llama-server -m _models/LFM2-24B-A2B-Q4_K_M.gguf --port 8080

# Any Ollama model (default port 11434)
ollama run llama3.2
```

--------------------------------

### Install Dependencies

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/cpt_translation_with_unsloth.ipynb

Installs necessary libraries including Unsloth, transformers, and other dependencies. It includes conditional installation for Google Colab environments.

```python
%%capture
import os, re
if "COLAB_" not in "".join(os.environ.keys()):
    !pip install unsloth
else:
    # Do this only in Colab notebooks! Otherwise use pip install unsloth
    import torch; v = re.match(r"[0-9]{1,}\\.[0-9]{1,}", str(torch.__version__)).group(0)
    xformers = "xformers==" + ("0.0.33.post1" if v=="2.9" else "0.0.32.post2" if v=="2.8" else "0.0.29.post3")
    !pip install --no-deps bitsandbytes accelerate {xformers} peft trl triton cut_cross_entropy unsloth_zoo
    !pip install sentencepiece protobuf "datasets==4.3.0" "huggingface_hub>=0.34.0" hf_transfer
    !pip install --no-deps unsloth
!pip install transformers==4.57.3
!pip install --no-deps trl==0.22.2
```

--------------------------------

### Colab/Kaggle Specific Installation

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/grpo_with_unsloth.ipynb

Handles installation for Colab/Kaggle environments, including conditional installation of libraries based on GPU type and specific versions of dependencies.

```python
#@title Colab Extra Install { display-mode: "form" }
%%capture
import os
!pip install --upgrade -qqq uv
if "COLAB_" not in "".join(os.environ.keys()):
    # If you're not in Colab, just use pip install!
    !pip install unsloth vllm
else:
    try: import numpy, PIL; get_numpy = f"numpy=={numpy.__version__}"; get_pil = f"pillow=={PIL.__version__}"
    except: get_numpy = "numpy"; get_pil = "pillow"
    try: import subprocess; is_t4 = "Tesla T4" in str(subprocess.check_output(["nvidia-smi"]))
    except: is_t4 = False
    get_vllm, get_triton = ("vllm==0.9.2", "triton==3.2.0") if is_t4 else ("vllm==0.11.2", "triton")
    !uv pip install -qqq --upgrade \
        unsloth {get_vllm} {get_numpy} {get_pil} torchvision bitsandbytes xformers
    !uv pip install -qqq {get_triton}
!uv pip install transformers==4.57.1
!uv pip install --no-deps trl==0.22.2
```

--------------------------------

### Build the Docker image

Source: https://github.com/liquid4all/cookbook/blob/main/examples/vl-webgpu-demo/README.md

Builds the Docker image for the application.

```sh
# Build the Docker image
docker build -t lfm-vl-webgpu -f LFM2.5-VL-1.6B-WebGPU/Dockerfile .
```

--------------------------------

### Install Dependencies

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/sft_for_vision_language_model.ipynb

Installs necessary libraries including Unsloth, transformers, and other dependencies. It includes conditional installation for Google Colab environments.

```python
%%capture
import os, re
if "COLAB_" not in "".join(os.environ.keys()):
    !pip install unsloth
else:
    # Do this only in Colab notebooks! Otherwise use pip install unsloth
    import torch; v = re.match(r"[0-9]{1,}\.[0-9]{1,}", str(torch.__version__)).group(0)
    xformers = "xformers==" + ("0.0.33.post1" if v=="2.9" else "0.0.32.post2" if v=="2.8" else "0.0.29.post3")
    !pip install --no-deps bitsandbytes accelerate {xformers} peft trl triton cut_cross_entropy unsloth_zoo
    !pip install sentencepiece protobuf "datasets==4.3.0" "huggingface_hub>=0.34.0" hf_transfer
    !pip install --no-deps unsloth
!pip install git+https://github.com/huggingface/transformers.git@3c2517727ce28a30f5044e01663ee204deb1cdbe
!pip install --no-deps trl==0.22.2
```

--------------------------------

### Deploy the server

Source: https://github.com/liquid4all/cookbook/blob/main/examples/voice-chat/README.md

Deploys the server component of the voice chat application to Modal.

```bash
make deploy-server
```

```bash
uv run modal deploy -m src.voice_chat.server
```

--------------------------------

### Run the container

Source: https://github.com/liquid4all/cookbook/blob/main/examples/vl-webgpu-demo/README.md

Runs the Docker container and exposes the application.

```sh
# Run the container
docker run -p 7860:7860 lfm-vl-webgpu
```

--------------------------------

### Install flash-attn

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/grpo_for_verifiable_tasks.ipynb

Installs the flash-attn library, which is often used for optimized attention mechanisms in deep learning models. The output indicates successful installation.

```bash
#!pip install trackio

  Created wheel for flash-attn: filename=flash_attn-2.8.3-cp312-cp312-linux_x86_64.whl size=253780426 sha256=4e2f9e39313266b1544b68138b15b91ee6221eccf14f7902b7c6620351340810
  Stored in directory: /root/.cache/pip/wheels/3d/59/46/f282c12c73dd4bb3c2e3fe199f1a0d0f8cec06df0cccfeee27
Successfully built flash-attn
Installing collected packages: flash-attn
Successfully installed flash-attn-2.8.3
```

--------------------------------

### Start the app server

Source: https://github.com/liquid4all/cookbook/blob/main/examples/home-assistant/README.md

Command to start the local Home Assistant application server using uv and uvicorn.

```bash
uv run uvicorn app.server:app --port 5173 --reload
```

--------------------------------

### LFM format example

Source: https://github.com/liquid4all/cookbook/blob/main/finetuning/notebooks/sft_with_unsloth.ipynb

Example of the LFM format for multi-turn conversations.

```text
<|startoftext|><|im_start|>user
Hello!<|im_end|>
<|im_start|>assistant
Hey there!<|im_end|>

```

--------------------------------

### Kick off training on Modal

Source: https://github.com/liquid4all/cookbook/blob/main/examples/home-assistant/README.md

Starts LoRA SFT training on Modal for 5 epochs using a specified configuration file.

```bash
cd leap-finetune
uv run leap-finetune ../finetune/configs/LFM2-350M.yaml
```

--------------------------------

### Install llama.cpp on Linux

Source: https://github.com/liquid4all/cookbook/blob/main/examples/invoice-parser/README.md

Installs llama.cpp by building from source on Linux.

```bash
git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build --config Release
# Add build/bin to your PATH
```

--------------------------------

### Start the llama.cpp server

Source: https://github.com/liquid4all/cookbook/blob/main/examples/local-coding-assistant/tasks/create_mvp/plan.md

Command to start the llama.cpp server, specifying the model file, context size, and port.

```bash
# Start the llama.cpp server with LFM2-24B-A2B Q4_0
./llama-server \
  --model /path/to/LFM2-24B-A2B-Q4_0.gguf \
  --ctx-size 8192 \
  --port 8080
```

--------------------------------

### Install llama.cpp on macOS

Source: https://github.com/liquid4all/cookbook/blob/main/examples/invoice-parser/README.md

Installs llama.cpp using Homebrew on macOS.

```bash
brew install llama.cpp
```

--------------------------------

### Install uv

Source: https://github.com/liquid4all/cookbook/blob/main/examples/flight-search-assistant/README.md

Command to check if uv is installed.

```bash
uv --version
```

--------------------------------

### main.py - CLI Entry Point

Source: https://github.com/liquid4all/cookbook/blob/main/examples/local-coding-assistant/tasks/create_mvp/plan.md

Provides a simple REPL for interacting with the agent, handling input, history, and graceful exits.

```bash
# Use Anthropic (validation)
export LCA_BACKEND=anthropic
export ANTHROPIC_API_KEY=sk-ant-...
lca
```

--------------------------------

### Run the client

Source: https://github.com/liquid4all/cookbook/blob/main/examples/voice-chat/README.md

Runs the client component of the voice chat application, which interacts with the deployed server.

```bash
make run
```

```bash
uv run modal run -m src.voice_chat.client
```