dphn/dolphin-2.9.1-yi-1.5-34b

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:May 18, 2024License:apache-2.0Architecture:Transformer0.1K Open Weights Cold

Dolphin 2.9.1 Yi 1.5 34B is a 34 billion parameter instruction-tuned causal language model developed by Eric Hartford, Lucas Atkins, Fernando Fernandes, and Cognitive Computations, based on the Yi-1.5-34B architecture. It achieves a 77.4 MMLU score and is fine-tuned for conversational, instruction-following, and coding tasks, including initial agentic abilities and function calling. The model is uncensored and trained with an 8k sequence length, despite the base model's 4k context, using a rope theta of 1,000,000.0.

Loading preview...

Dolphin 2.9.1 Yi 1.5 34B: An Uncensored, High-Performance LLM

Dolphin 2.9.1 Yi 1.5 34B is a powerful 34 billion parameter language model developed by Eric Hartford, Lucas Atkins, Fernando Fernandes, and Cognitive Computations. Built upon the Yi-1.5-34B base, this model stands out for its impressive performance and unique characteristics.

Key Capabilities & Features

  • High MMLU Score: Achieves a notable 77.4 MMLU score, indicating strong general knowledge and reasoning abilities.
  • Extended Context Handling: While the base model has a 4k context, Dolphin 2.9.1 was fine-tuned with an 8k sequence length using a rope theta of 1,000,000.0, enhancing its ability to process longer inputs.
  • Versatile Skills: Excels in instruction following, conversational AI, and coding tasks. It also incorporates initial agentic abilities and supports function calling.
  • Uncensored Nature: The model is intentionally uncensored, designed to be highly compliant with user requests, including potentially unethical ones. Users are advised to implement their own alignment layers for responsible deployment.
  • ChatML Format: Utilizes the ChatML prompt template for structured interactions.
  • Training: Fine-tuned using a diverse dataset including ShareGPT, coder-specific datasets, Orca-Math, and agentic instruction data, with training performed using Axolotl.

Intended Use Cases

  • General Purpose AI Assistant: Capable of handling a wide range of conversational and instruction-based tasks.
  • Coding and Development: Suitable for code generation and related programming tasks.
  • Agentic Applications: Its initial agentic abilities and function calling support make it suitable for developing more complex, autonomous AI systems.
  • Research and Experimentation: Its uncensored nature offers flexibility for research into model behavior and alignment strategies.

Finetunes
1 models