mzbac/CodeLlama-34b-guanaco
The mzbac/CodeLlama-34b-guanaco is a 34 billion parameter CodeLlama base model fine-tuned on the text chunk from the OpenAssistant-Guanaco dataset. This model is specifically adapted for conversational tasks, though it may require a stop string like "### Human:" to manage response length. Its primary strength lies in generating human-like text based on the provided prompt template.
Loading preview...
mzbac/CodeLlama-34b-guanaco Overview
This model is a 34 billion parameter variant of the CodeLlama base model, specifically fine-tuned using a unique approach. Instead of traditional Q&A pairs, it was trained on the text chunks from the OpenAssistant-Guanaco dataset. This training methodology influences its conversational style and response generation.
Key Characteristics
- Base Model: CodeLlama 34B.
- Fine-tuning Data: OpenAssistant-Guanaco dataset (text chunks).
- Context Length: Supports a context window of 32768 tokens.
- Response Management: Due to its fine-tuning on text chunks, the model may struggle to determine the end of an answer. It is recommended to use a stop string, such as
"### Human:", to prevent the model from generating excessively long or self-conversational responses.
Recommended Usage
This model is suitable for conversational AI applications where the user can implement a stop string to control output. It follows a simple prompt template:
### Human: {prompt}
### Assistant:Users should be prepared to manage the model's output length for optimal performance in interactive scenarios.