metadata
license: mit
datasets:
- CreitinGameplays/merged-data-v2
base_model:
- HuggingFaceH4/zephyr-7b-beta
- mistral-community/Mistral-7B-v0.2
language:
- en
ConvAI-9b: A Conversational AI Model
1. Model Details
- Model Name: ConvAI-9b
- Authors: CreitinGameplays
- Date: April 18th, 2024
2. Model Description
ConvAI-9b is a fine-tuned conversational AI model with 9 billion parameters. It is based on the following models:
- Base Model: HuggingFaceH4/zephyr-7b-beta
- Merged Model: mistral-community/Mistral-7B-v0.2
3. Training Data
The model was fine-tuned on a custom dataset of conversations between an AI assistant and a user. The dataset format followed a specific structure:
<|system|> (system prompt, e.g.: You are a helpful AI language model called ChatGPT, your goal is helping users with their questions) </s> <|user|> (user prompt) </s>
4. Intended Uses
ConvAI-9b is intended for use in conversational AI applications, such as:
- Chatbots
- Virtual assistants
- Interactive storytelling
- Educational tools
5. Limitations
- Like any other language model, ConvAI-9b may generate incorrect or misleading responses.
- It may exhibit biases present in the training data.
- The model's performance can be affected by the quality and format of the input text.
6. Evaluation
Metrics | Value |
---|---|
ARC | 57.50 |
HellaSwag | 80.34 |
TruthfulQA | 49.54 |
Winogrande | 76.24 |
More detailed evaluation here