google/gemma_pytorch
Gemma Pytorch
The official PyTorch implementation of Google's Gemma models
Features
- Long-Context Models - Supports input sequences up to 256K tokens, enabling reasoning over large documents or conversations.
- Long Context Processing - Supports input sequences up to 256K tokens for reasoning over large documents or conversations.
- Text-Only Inference Engines - Loads and executes a decoder-only language model on CPU, GPU, or TPU for generating text completions.
- Multimodal Model Runners - Ships a PyTorch implementation that loads and runs Gemma models for multimodal text-and-image inference.
- Pretrained Model Loading - Downloads and loads open-weight model checkpoints from Kaggle or Hugging Face for local inference.
- PyTorch Implementations - Provides an official PyTorch port of Google's Gemma language models for inference and fine-tuning.
- LLM Inference Engines - Runs Google's Gemma models for text generation using PyTorch on CPU, GPU, or TPU.
- Pretrained Checkpoint Loaders - Downloads model weights and tokenizer files from Kaggle or Hugging Face Hub for any supported variant.
- Open-Weight Checkpoint Loaders - Downloads and loads pretrained Gemma model weights from Kaggle or Hugging Face.
- Decoder-Only Inference - Loads and executes a decoder-only transformer to generate text completions from a prompt on CPU, GPU, or TPU.
- Model Fine-Tuning - Customizes pre-trained Gemma models on specific datasets to improve task performance.
- Multimodal Input Processors - Accepts text, audio, and image data as input to a single model for combined understanding and generation.
- AI Safety Guardrails - Includes a safety evaluation module that checks model inputs and outputs against defined safety policies.
- Function Calling Interfaces - Generates structured function-call arguments from natural language prompts for agentic workflows.
- Content Safety Classifiers - Implements a safety classifier that evaluates text against policy-defined categories to detect violations.
- Text Embedding Generators - Produces dense vector representations of text for search, clustering, or similarity tasks.
- Text Embeddings - Produces numerical vector representations of text for use in retrieval, similarity search, and classification.
- Model Safety Filters - Evaluates model inputs and outputs against predefined safety policies to detect violations.
- Large Language Models (LLMs) - Listed in the “Large Language Models (LLMs)” section of the The Incredible Pytorch awesome list.
