gemma3-torchao-quant-sparse

A performance-optimized adaptation of the google/gemma-3-4b-it multimodal LLM, which integrates memory and compute efficiency techniques while preserving high-quality generative capabilities for image-text tasks.

Key Features

1. INT8 Weight-Only Quantization

• Uses torchao’s Int8WeightOnlyConfig for weight-only INT8 quantization.
• Reduces VRAM usage significantly and speeds up inference.
• Maintains model output fidelity while lowering memory footprint.

2. Sparsity Techniques

2.1 Magnitude-Based Pruning

• Zeroes out the smallest weights in linear layers based on absolute magnitude.
• Safely reduces parameter count without altering layer shapes.

2.2 Gradual Pruning

• Incrementally increases sparsity over multiple steps (e.g., 1000 steps).
• Prevents sudden degradation in output quality.
• Ideal for safe exploration of sparsity ratios.

2.3 Layer-Norm-Based Pruning

• Prunes weights in linear layers with lowest L2 norms per layer.
• Extremely fast because it does not require forward passes.
• Preserves output quality and keeps layer shapes intact, ensuring seamless integration with the model.

2.4 Flexible Sparsity Ratios

• Supports low ratios (1–2%) for minimal impact and ratios (up to 40%) for optimization.

2.5 Filter Maps for Quantization and Pruning

• Excludes critical layers such as embeddings, normalization layers and output heads from pruning and quantization.
• Ensures that pruning does not break model outputs and preserves output quality.

Benefits

• Significantly lower VRAM usage (~5–6 GB for 4B model at 500–750 tokens).
• Faster inference with Torch Compile + INT8 quantization.
• Safe sparsity techniques allow memory-efficient experiments without breaking the model.
• Maintains high-quality multimodal generation (image-to-text and text-to-text).