---
base_model: NousResearch/Nous-Hermes-2-Yi-34B
datasets:
- gbstox/agronomy-resources
tags:
- Yi-34B
- instruct
- finetune
- agriculture
language:
- en
---

# AgronomYi-hermes-34B

<img src="https://cdn-uploads.huggingface.co/production/uploads/63042a3d7373aacccd896484/TwXNxFw8zSLuWjiYL41Bj.jpeg" width="500" >

# About
AgronomYi is a fine tune of [Nous-Hermes-2-Yi-34B](https://huggingface.co/NousResearch/Nous-Hermes-2-Yi-34B), which uses Yi-34B as the base model.
I fine tuned this with agronomy data (exclusively textbooks & university extension guides), full training data set [here](https://huggingface.co/datasets/gbstox/agronomy-resources)).
AgronomYi outperforms all models on the benchmark except for gpt-4, and consistently beats the base model by 7-9% and the hermes fine tune by 3-5%. I take this to mean that even better results can be acheived with additional fine tuning, and larger models tend to perform better in general. 

# Benchmark comparison
[benchmark info here](https://github.com/gbstox/agronomy_llm_benchmarking)

| Model Name | Score | Date Tested |
|------------|-------|-------------|
| gpt-4 | 85.71% | 2024-01-15 |
| agronomYi-hermes-34b | 79.05% | 2024-01-15 |
| mistral-medium | 77.14% | 2024-01-15 |
| nous-hermes-yi-34b | 76.19% | 2024-01-15 |
| mixtral-8x7b-instruct | 72.38% | 2024-01-15 |
| claude-2 | 72.38% | 2024-01-15 |
| yi-34b-chat | 71.43% | 2024-01-15 |
| norm | 69.52% | 2024-01-17 |
| openhermes-2.5-mistral-7b | 69.52% | 2024-01-15 |
| gpt-3.5-turbo | 67.62% | 2024-01-15 |
| mistral-7b-instruct | 61.9% | 2024-01-15 |