README.md

metadata

datasets:
  - Delta-Vector/Orion-Misc-Data-Sharegpt-Prefixed
  - Delta-Vector/Orion-Basket-Weaving-Filtered
  - Delta-Vector/Orion-vanilla-backrooms-claude-sharegpt
  - Delta-Vector/Orion-Roleplay-Logs-Sharegpt-Ngram-cleaned
  - Delta-Vector/Orion-BlueSky-10K-Complexity
base_model:
  - zerofata/L3.3-GeneticLemonade-Unleashed-v3-70B
tags:
  - roleplay
  - chat
  - creative-writing

Shimamura 70B

Created by Delta-Vector →

Model Information

Shimamura-70B

70B parameters Creative / Funny Chat Model

This is a Finetune of zerofata/L3.3-GeneticLemonade-Unleashed-v3-70B to be a good Chat Model at a larger parameter size

This model has been trained on 100M tokens of Human chat logs from Bsky, 4chan & Most of all ShoujoAI.

Support me on Ko-Fi: https://ko-fi.com/deltavector

Quantized Versions

Available Downloads

• GGUF FormatFor use with LLama.cpp & Forks(Coming Soon!)
• EXL2 FormatFor use with TabbyAPI (Coming Soon!)
• EXL3 FormatFor use with TabbyAPI (Slower on Ampere))

Prompting

Model has been tuned with the Llama-3-Instruct formatting.

Samplers

For testing of this model, I used Temp=1, 0.1 Min-P.

See Axolotl Config

            https://wandb.ai/new-eden/austral/artifacts/axolotl-config/config-c61un0ze/v0/files/axolotl_config_cu4t7u4q.yml
            

Training

The training was done for 4 epoch using 8 x H200s GPUs for the fine-tuning of the model.

Credits

Thank you to Lucy Knada, Zerofata, Auri, Intervitens, Cgato, Kubernetes Bad and the rest of Anthracite.