wenbopan/Faro-Yi-34B

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:Apr 1, 2024License:mitArchitecture:Transformer0.0K Open Weights Cold

Faro-Yi-34B is a 34 billion parameter chat model developed by wenbopan, based on the Yi-34B-200K architecture. It is extensively instruction-tuned on the Fusang-V1 dataset, enhancing its performance across various downstream tasks and long-context modeling. This model supports a 200K context length and excels at handling complex instructions and lengthy documents in both English and Chinese.

Loading preview...

Overview

Faro-Yi-34B is an instruction-tuned chat model developed by wenbopan, building upon the Yi-34B-200K base model. It features 34 billion parameters and supports an impressive 200K context length, making it highly capable for processing extensive inputs.

Key Capabilities

  • Enhanced Task Performance: Significantly improved capabilities across diverse downstream tasks compared to its base model.
  • Robust Long-Context Modeling: Delivers stable and reliable results even with lengthy documents or complex, multi-turn instructions.
  • Bilingual Support: Seamlessly operates in both English and Chinese environments.
  • Practical Application Focus: Designed for practicality, ensuring higher quality and reliable outputs for real-world use cases.

Training and Differentiators

Faro-Yi-34B's enhanced performance stems from extensive instruction tuning on the large-scale synthetic dataset, Fusang-V1. This training methodology has equipped the model with superior handling of complex queries and long-form content, distinguishing it from other models by its focus on practical, high-quality output in challenging scenarios.

Good For

  • Applications requiring processing and understanding of very long documents.
  • Chatbots or assistants needing to maintain context over extended conversations.
  • Tasks involving complex instructions or multi-step reasoning in English or Chinese.