dolphin-2_2-yi-34b

Maintained By
cognitivecomputations

Dolphin-2.2-Yi-34B

PropertyValue
Base ModelYi-34B
Context Length16k tokens
LicenseApache 2.0
Training Duration3 days on 4x A100s

What is dolphin-2_2-yi-34b?

Dolphin-2.2-Yi-34B is an advanced language model built on the Yi-34B architecture, specifically designed for enhanced conversational abilities and empathetic responses. Sponsored by a16z, this model represents a significant evolution in the Dolphin series, incorporating sophisticated training on multiple high-quality datasets including Dolphin, Airoboros, Samantha, and WizardLM.

Implementation Details

The model utilizes the ChatML prompt format and was trained using qLoRA and Axolotl frameworks. It's built on a llama-compatible version of Yi-34B, featuring uncensored capabilities while maintaining high compliance with user requests.

  • Trained for 3 epochs using advanced quantization techniques
  • Implements 16k context window for extended conversations
  • Uses specialized dataset combinations for enhanced performance
  • Supports multi-turn conversations with improved context understanding

Core Capabilities

  • Enhanced conversational abilities with natural dialogue flow
  • Improved empathy and emotional understanding
  • Extended context handling for complex discussions
  • Uncensored responses with high compliance
  • Sophisticated personal advice capabilities

Frequently Asked Questions

Q: What makes this model unique?

This model stands out for its combination of conversational abilities and empathy, achieved through careful dataset curation and training methodology. The integration of Samantha and WizardLM data specifically enhances its ability to maintain meaningful multi-turn conversations and provide emotionally intelligent responses.

Q: What are the recommended use cases?

The model is particularly well-suited for applications requiring extended conversations, personal advice, and situations where emotional intelligence is valuable. However, due to its uncensored nature, implementing appropriate alignment layers is recommended before deployment in production environments.

The first platform built for prompt engineering