Hello, I have a model that works well for my assistant. I want to finetune this model - however, given that this can get a bit pricey, I don’t want to waste money on doing that only to find out I won’t be able to select it as one of the options for my assistant.
So, is it possible to use a finetuned model for an assistant?
Hi,
If you would like to fine-tune a reasonably sized model at a lower cost, especially from a notebook on Google Colab, I invite you to take a look at how to use Unsloth.ai.
From the README.md :
| Unsloth supports | Free Notebooks | Performance | Memory use |
|---|---|---|---|
| Llama 3 (8B) | Start for free | 2x faster | 60% less |
| Mistral v0.3 (7B) | Start for free | 2.2x faster | 73% less |
| Phi-3 (medium) | Start for free | 2x faster | 50% less |
| Phi-3 (mini) | Start for free | 2x faster | 50% less |
| Gemma (7B) | Start for free | 2.4x faster | 71% less |
| ORPO | Start for free | 1.9x faster | 43% less |
| DPO Zephyr | Start for free | 1.9x faster | 43% less |
| TinyLlama | Start for free | 3.9x faster | 74% less |
Link to GitHub : GitHub - unslothai/unsloth: Finetune Llama 3, Mistral, Phi & Gemma LLMs 2-5x faster with 80% less memory