Yi-34B-Chat API

Description

Designed as a bilingual model and trained on a 3T multilingual corpus, the Yi series models rank among the top LLMs globally, excelling in language comprehension, commonsense reasoning, reading comprehension, and beyond. Try out at our AI API Playground.

What are Use Cases for Yi-34B

Yi-34B is primarily utilized for customer service chatbots and interactions, with its fine-tuned capabilities extending to technical support, educational tutoring, and creative writing support across various industries.

How does it compare to Mixtral and ChatGPT?

Yi-34B is primarily utilized for customer service chatbots and interactions, with its fine-tuned capabilities extending to technical support, educational tutoring, and creative writing support across various industries.

Tips

• Temperature Setting: Yi-34B tends to run hot by default. It's recommended to adjust the temperature setting to around 0.5 to manage this.
• Repetition Penalty: Set the repetition penalty between 1.1 and 1.2 to avoid repetitive outputs.
• Minimum Probability (min_p): To control uncreative one-sentence responses, set min_p between 0.05 and 0.1.
• Instruction Template: The model responds well to the "ChatML" instruction template.
• Custom Stopping String: If the model outputs <s> at the end of its answers, use <s> as a custom stopping string.
• Beginning of Sentence (bos) Token: Some Yi models perform better when the "Add the bos_token to the beginning of prompts" option is unchecked.

Use cases

• Customer service chatbots: Powering chatbots that respond to customer inquiries and provide support.
• Technical support: Helping users with technical problems through efficient and accurate responses.
• Educational tutoring: Providing personalized tutoring and answering educational queries.
• Creative writing support: Assisting in content creation, story generation, and other creative writing tasks.
• Conversational dialogue: Engaging in multi-turn dialogue for natural and fluent interactions.
• Language translation: Translating text between multiple languages, including Chinese and English.
• Coding assistance: Offering code generation, explanation, and debugging help in programming languages.
• Multilingual virtual agents: Serving global audiences with high-accuracy conversational AI across languages.
• AI-driven knowledge systems: Building context-aware, instruction-following agents for knowledge search and data retrieval.
• Enterprise applications: Scalable deployments for cloud, on-prem, or edge AI in customer service, content generation, and virtual agents.

Capabilities

• Function calling
• Streaming
• Structured outputs