In February we launched LoRA Land, a collection of 25+ fine-tuned Mistral-7b models that outperform or rival GPT-4 on specific tasks. All of these models were trained for less than $8 each and served on a single GPU with Predibase and open-source LoRAX.
Since then, we’ve fine-tuned popular open-source LLMs, Gemma, Phi, Llama, and Zephyr, on the same 25+ datasets to provide detailed benchmarks on which models perform best across tasks. Join our upcoming webinar to dive deep into the results and our methodology, and learn how you can efficiently fine-tune and serve your own LLMs that are on par with GPT-4.
In this on-demand webinar, Staff Software Engineer Justin Zhao, and ML Engineer Timothy Wang, lead an in-depth discussion on our findings:
• How we fine-tuned open-source LLMs that rival GPT-4
• How you can implement Parameter-Efficient Fine-Tuning (PEFT) methods like Low Rank Adaptation (LoRA)
• Which tasks are best suited for fine-tuning based on our benchmarks
• Which popular LLMs—namely Phi, Gemma, and Mistral—perform best and worst across tasks
• How we implemented an evaluation harness for fine-tuning at scale
By the end of this session, you will be able to use our framework to cost-effectively build your own production LLM applications.
Ready to get started?
• Download the webinar slides:
• Access our evaluation harness:
• Visit LoRA Land to prompt our 25 open-source adapters:
• Efficiently fine-tune and serve your own LLMs with $25 in free Predibase:
[ad_2]
source
