Tinyllama 1.1B Chat V0.6
TinyLlama is a 1.1 billion parameter Llama model pre-trained on 3 trillion tokens, suitable for scenarios with limited computation and memory resources.
Downloads 11.60k
Release Time : 11/20/2023
Model Overview
TinyLlama is a compact large language model that fully replicates the architecture and tokenizer of Llama 2, compatible with numerous Llama-based open-source projects.
Model Features
Efficient training
Pre-trained on 3 trillion tokens in just 90 days using only 16 A100-40G GPUs.
Compatibility
Fully replicates the Llama 2 architecture and tokenizer, compatible with Llama-based open-source projects.
Fine-tuning approach
Adopts HuggingFace Zephyr's training scheme, including UltraChat fine-tuning and DPO alignment.
Model Capabilities
Text generation
Dialogue systems
Use Cases
Chatbots
Stylized chatbots
Can customize chatbots with different styles, such as pirate-themed.
Generates dialogue content that matches the set style.
Resource-constrained applications
Edge device deployment
Suitable for running on devices with limited computation and memory resources.
Featured Recommended AI Models
Š 2025AIbase