scb10x/llama-3-typhoon-v1.5-8b-instruct

Warm
Public
8B
FP8
8192
1
May 6, 2024
License: llama3
Hugging Face
Overview

Llama-3-Typhoon-v1.5-8B-Instruct Overview

Llama-3-Typhoon-v1.5-8B-Instruct is an 8 billion parameter instruction-tuned decoder-only model developed by SCB 10X, built upon the Llama 3 foundation model. It is designed to excel in both Thai and English language processing, with a particular focus on improving performance for Thai-specific tasks.

Key Capabilities & Performance

  • Bilingual Support: Primarily supports Thai (🇹🇭) and English (🇬🇧) languages.
  • Enhanced Thai Performance: Demonstrates significant improvements over its predecessor (Typhoon-1.0) and other comparable models on various Thai examination benchmarks (ONET, IC, TGAT, TPAT-1, A-Level), achieving an average score of 0.506.
  • General Instruction Following: As an instruct model, it is fine-tuned to follow user instructions effectively.
  • Llama 3 Architecture: Leverages the robust Llama 3 architecture, ensuring a strong base for its capabilities.

Intended Uses & Limitations

This model is suitable for a wide range of instruction-based applications in both Thai and English. However, as an evolving model, users should be aware that it may occasionally produce inaccurate, biased, or objectionable content. Developers are advised to assess these risks within their specific use cases.