TheBloke/stable-vicuna-13B-HF
TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Apr 28, 2023License:cc-by-nc-sa-4.0Architecture:Transformer0.1K Open Weights Cold
TheBloke/stable-vicuna-13B-HF is an unquantized float16 model of CarperAI's StableVicuna 13B, a 13 billion parameter language model based on the LLaMA transformer architecture. Fine-tuned using Reinforcement Learning from Human Feedback (RLHF) via Proximal Policy Optimization (PPO), it excels at conversational and instructional tasks. This model is intended for text generation with a focus on conversational applications, offering a robust foundation for further fine-tuning.
Loading preview...
StableVicuna-13B: RLHF-Tuned Conversational Model
This model, provided by TheBloke, is an unquantized float16 version of CarperAI's StableVicuna-13B. It is a 13 billion parameter auto-regressive language model built upon the LLaMA transformer architecture.
Key Capabilities
- RLHF Fine-tuning: Enhanced for conversational and instructional tasks through Reinforcement Learning from Human Feedback (RLHF) using Proximal Policy Optimization (PPO).
- Diverse Training Data: Fine-tuned on a mix of datasets including OpenAssistant Conversations Dataset (OASST1), GPT4All Prompt Generations, and Alpaca, focusing on assistant-style conversations and instructions.
- Llama Base: Merges deltas from CarperAI's StableVicuna 13B with original Llama 13B weights.
Good For
- Conversational AI: Designed for text generation in conversational contexts.
- Instruction Following: Capable of responding to instructions effectively due to its training on instruction-based datasets.
- Further Fine-tuning: Serves as a strong base model for users looking to fine-tune on their specific data for improved performance on particular tasks.