### Install SNAC-Go Package

Source: https://github.com/tcpipuk/snac-go/blob/main/README.md

Use 'go get' to install the SNAC-Go library. This command fetches and installs the package and its dependencies.

```bash
go get github.com/tcpipuk/snac-go
```

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

### Install Dependencies and Run Conversion Inside Docker

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Set up a Python 3.13 Docker container, install necessary packages including the SNAC library, and then run the weight conversion script.

```bash
docker run --rm -it -v $(pwd):/workspace -w /workspace python:3.13 bash

# Inside container:
pip install torch numpy
git clone https://github.com/hubertsiuzdak/snac tmp/snac
cd tmp/snac && pip install -e . && cd ../..

# Run conversion
python scripts/convert_weights.py hubertsiuzdak/snac_24khz -o weights/snac_24khz.npz
```

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

### Install Local Python Dependencies

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Install project dependencies from requirements.txt and set up the local SNAC package for use with the conversion script.

```bash
pip install -r scripts/requirements.txt
git clone https://github.com/hubertsiuzdak/snac tmp/snac
cd tmp/snac && pip install -e . && cd ../..
```

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

### Install PyTorch and NumPy

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Install PyTorch and NumPy if they are missing. This addresses the 'No module named 'torch'' error.

```bash
pip install torch numpy
```

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

### Install SNAC Package

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Clone the SNAC repository and install it locally. This is a solution for when the SNAC package is not found.

```bash
git clone https://github.com/hubertsiuzdak/snac tmp/snac
cd tmp/snac && pip install -e . && cd ../..
```

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

### Quick Start: Decode SNAC Tokens to Audio

Source: https://github.com/tcpipuk/snac-go/blob/main/README.md

This Go snippet demonstrates loading the SNAC decoder and decoding tokens into an audio waveform. Ensure 'tokens' is correctly formatted and 'writeWAV' function is available.

```go
package main

import (
    "github.com/tcpipuk/snac-go/snac"
    "log"
)

func main() {
    // Load SNAC decoder with 24kHz model
    decoder, err := snac.NewDecoder("hubertsiuzdak/snac_24khz")
    if err != nil {
        log.Fatal(err)
    }

    // Decode SNAC tokens to audio
    // tokens is [3][]int representing 3 hierarchical codebook levels
    audio, err := decoder.Decode(tokens)
    if err != nil {
        log.Fatal(err)
    }

    // Write to WAV file
    if err := writeWAV("output.wav", audio, 24000); err != nil {
        log.Fatal(err)
    }
}
```

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

### Test Mode Validation Output

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

This example shows the expected output when running the conversion script in test mode, demonstrating successful validation without saving.

```bash
python scripts/convert_weights.py hubertsiuzdak/snac_24khz --test

# Expected output:
#   Loading model from: hubertsiuzdak/snac_24khz
#   Extracted 135 parameters:
#     - Quantizer: 20
#     - Decoder: 115
#   Validating weights...
#     ✓ Found 67 weight_g parameters
#     ✓ All quantizer codebooks present
#     ✓ Decoder structure valid
#     ✓ All dtypes are float32
#     ✓ Weight normalisation pairs complete
#   ✓ Test mode: Validation passed, no file saved
```

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

### Run Weight Conversion in Docker (Custom)

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Execute the Docker wrapper script with custom model and output path arguments. This allows conversion of different models and saving to specified locations.

```bash
./scripts/convert_in_docker.sh hubertsiuzdak/snac_32khz weights/snac_32khz.npz
```

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

### Convert Weights with Verbose Output

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Execute the weight conversion script with verbose output enabled to see detailed progress and information during the process. Use -o for output path.

```bash
python scripts/convert_weights.py hubertsiuzdak/snac_24khz -o weights/snac_24khz.npz --verbose
```

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

### Convert Weights from Local Path

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Convert model weights from a local directory to a .npz file. Specify the source directory and the desired output file name.

```bash
python scripts/convert_weights.py /path/to/local/model -o model.npz
```

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

### Run Conversion Script in Docker

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Execute the weight conversion script within a Docker container. This is the recommended method on TrueNAS systems.

```bash
./scripts/convert_in_docker.sh
```

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

### Weight Normalization Logic in Go

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

This Go code snippet demonstrates how weight normalization parameters (g and v) are handled. It shows the L2 norm calculation per output channel and the subsequent folding of normalized weights. This approach aids in verifying Python reference logic and debugging numerical differences.

```go
// Folded in Go:
// w = g * (v / ||v||)
v_norm := computeL2Norm(v, axis=0)  // Per output channel
w := g * (v / (v_norm + 1e-9))      // Broadcast g appropriately
```

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

### Load SNAC Decoder in Go

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Load a pre-converted SNAC decoder using its weights file in Go. Handles potential errors during loading.

```go
import "github.com/tcpipuk/snac-go/snac"

// Load decoder with converted weights
decoder, err := snac.NewDecoder("weights/snac_24khz.npz")
if err != nil {
    log.Fatal(err)
}

// Decode tokens to audio
audio, err := decoder.Decode(tokens)
```

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

### Regenerate SNAC Test Data with Docker

Source: https://github.com/tcpipuk/snac-go/blob/main/snac/testdata/README.md

Use this command to regenerate all test data using Docker. Ensure you are in the project root directory.

```bash
docker run --rm -v $(pwd):/workspace -w /workspace \
  python:3.13 bash -c \
    "pip install torch numpy huggingface_hub && \
    cd tmp/snac && pip install -e . && cd ../.. && \
    python scripts/generate_ground_truth.py"

```

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

### Convert SNAC PyTorch Model to NumPy

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Convert a SNAC PyTorch model from HuggingFace to a compressed NumPy .npz file. Specify the model name and output file path.

```python
python scripts/convert_weights.py hubertsiuzdak/snac_24khz -o weights/snac_24khz.npz
```

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

### Test SNAC Model Conversion

Source: https://github.com/tcpipuk/snac-go/blob/main/scripts/README.md

Run the weight conversion script in test mode to validate the output structure and completeness without saving the .npz file.

```python
python scripts/convert_weights.py hubertsiuzdak/snac_24khz --test
```

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

### Regenerate SNAC Test Data Directly

Source: https://github.com/tcpipuk/snac-go/blob/main/snac/testdata/README.md

Run this command directly if you have a Python environment available to regenerate test data.

```bash
python scripts/generate_ground_truth.py

```

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

### SNAC 24kHz Model Configuration

Source: https://github.com/tcpipuk/snac-go/blob/main/docs/architecture.md

Configuration parameters for the 24kHz speech-optimised SNAC model. These settings define the model's architecture and capacity.

```json
{
  "sampling_rate": 24000,
  "encoder_dim": 64,
  "encoder_rates": [3, 3, 5, 5],
  "latent_dim": 64,
  "decoder_dim": 1536,
  "decoder_rates": [5, 5, 3, 3],
  "attn_window_size": 32,
  "codebook_size": 4096,
  "codebook_dim": 8,
  "vq_strides": [8, 4, 2, 1]
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.