Nicholai/jan
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add generating training set section

Browse Source
This commit is contained in:
hahuyhoang411 2024-03-01 13:30:48 +07:00
parent ce0c427f57
commit ed3bf86d84
1 changed files with 9 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
docs/blog/rag-is-not-enough.md
View File

@ -77,4 +77,12 @@ So, we directed our efforts toward training a model to answer user questions bas
Specifically, we trained it on Nitro [docs](https://nitro.jan.ai/docs). For context, Nitro is the default inference engine for Jan. It’s a serious server implementation of LlamaCPP, written in C++, with multimodal, queues, and other production-level server capabilities. Specifically, we trained it on Nitro [docs](https://nitro.jan.ai/docs). For context, Nitro is the default inference engine for Jan. It’s a serious server implementation of LlamaCPP, written in C++, with multimodal, queues, and other production-level server capabilities.
It made an interesting corpus because it was rife with post-2023 technical jargon, edge cases, and poor informational layout. It made an interesting corpus because it was rife with post-2023 technical jargon, edge cases, and poor informational layout.
## Generating a training dataset for GPT-4 and training
The first step was to transform Nitro’s unstructured format into a synthetic Q&A dataset designed for [instruction tuning](https://arxiv.org/pdf/2109.01652.pdf).
The text was split into chunks of 300-token segments with 30-token overlaps. This was to target a GPT-4 with 8k context length. This helped to avoid a [lost-in-the-middle](https://arxiv.org/abs/2307.03172) problem where LLM can’t use context efficiently to answer given questions.
The chunks were then given to **GPT-4** to generate 3800 Q&A pairs. You can find the [open-sourced dataset here](https://huggingface.co/datasets/jan-hq/nitro_binarized_v2) on HuggingFace.