ignos/LeoScorpius-GreenNode-Alpaca-7B-v1

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Dec 15, 2023License:apache-2.0Architecture:Transformer Open Weights Cold

ignos/LeoScorpius-GreenNode-Alpaca-7B-v1 is a 7 billion parameter Mistral-based language model developed by Ignos. It is a merge of viethq188/LeoScorpius-7B-Chat-DPO and GreenNode/GreenNodeLM-7B-v1olet, further fine-tuned on the tatsu-lab/alpaca dataset. This model is designed for comparative analysis of behaviors and metrics against its base model and other Mistral-based models fine-tuned on different datasets, making it suitable for research and evaluation purposes.

Loading preview...

Model Overview

ignos/LeoScorpius-GreenNode-Alpaca-7B-v1 is a 7 billion parameter language model developed by Ignos, built upon the Mistral architecture. This model is a unique fusion, created by merging two existing models: viethq188/LeoScorpius-7B-Chat-DPO and GreenNode/GreenNodeLM-7B-v1olet. Following this merge, the model underwent additional fine-tuning using the tatsu-lab/alpaca dataset.

Key Characteristics

  • Architecture: Based on the Mistral-7B-v0.1 model.
  • Development: Developed by Ignos, with a focus on comparative analysis.
  • Training Method: Utilizes a QLoRA approach for fine-tuning, followed by merging with the base model.
  • License: Released under the Apache-2.0 license.

Intended Use Cases

This model is primarily designed for research and evaluation. Its creation facilitates:

  • Behavioral Comparison: Analyzing and comparing its responses and characteristics against the original Mistral base model.
  • Metric Evaluation: Benchmarking its performance and metrics against other Mistral-based models that have been fine-tuned on different datasets.

Technical Details

The training was conducted on RunPod, utilizing 4 x Nvidia RTX 4090 GPUs, 64 vCPUs, and 503 GB RAM. Software used included Mergekit and Axolotl 0.3.0, with PEFT 0.6.0. Quantization during training employed bitsandbytes with bnb_4bit_quant_type: nf4 and bnb_4bit_use_double_quant: True.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p