Custom GPT: Finetune LLaMa 2 by PEFT


By David Nordström

2 Jun 2023



In this blog post I dive into the plans for ChatterFlow to finetune our own proprietary model based on the open sourced weights available today.


After the publication of the weights of Meta's large scale language model (LLM) LLaMA 2, there has been a surge in the capabilities of cheap finetuning. Through ūü§ó HuggingFace's PEFT library, users have been enabled to finetune LLMs with consumer hardware. The results have been staggering, for specific tasks these simpler LLMs outperform the likes of¬†GPT4¬†when tuned on custom datasets.

Collecting user data through our interactions, we can leverage internal data to finetune an LLM. However, LLaMA 2 has limited Swedish knowledge. Once this is provided, or I assemble a team that can train our own pre-trained model from scratch, I must rely on a sub-par Swedish model in LLaMA 2 or pay for finetuning of GPT4.


There is a large opportunity to create a pipeline, leveraging commercially licensed models as they are published, to capture the non-english speaking information services markets. Though, we stand at a crossroads, either we train our own model from scratch or we wait for the open-sourced LLMs to improve and deploy finetuning.

  • Low hardware requirement to deploy strong finetuned model.
  • Commercial mote through access to Swedish data and queries.
  • Commercially viable open source models have been published (LLaMA 2)
  • Meanwhile, one can finetune OpenAI's models, but to quite a high cost.

LLMs provide a unique business opportunity of which we are at the fore-front. Join me in seizing this opportunity and building something special.

With this said, I believe we can have the joint strength to capture this LLM opportunity. Do not hesitate to reach out to us and join the journey!

Popular Tags :
Share this post :