Maybe this helps people understand what they are doing at index time.
* Version 1. Ask the LLM to describe the code snippet. Create an embedding of the description. LLM generation + embeddings required.
* Version 2. run the code snippet directly through the embedding API. Skip the LLM text generations step. Now run the code snippet through the bias matrix and finally index the resulting embedding.
I assume this only works b/c they fine tuned a bias matrix on code snippet and text pairs. Feels more like a light version of transfer learning to me.
The article was a little unclear in the actual approach for V2 so if I have anything wrong please correct me.
I haven't had the same issues with my framework. I'm running an Intel 11-gen i5 version of a framework I bought in Feb of '22. Sure the battery life sucks for Fedora and the trackpad is a overly sensitive but overall it's worked fine for me. I complained to them about a noisy fan and they sent me a new one.
Perhaps there's some variability in the build quality between machines or batches. I've been happy with my purchase.
* Version 1. Ask the LLM to describe the code snippet. Create an embedding of the description. LLM generation + embeddings required.
* Version 2. run the code snippet directly through the embedding API. Skip the LLM text generations step. Now run the code snippet through the bias matrix and finally index the resulting embedding.
I assume this only works b/c they fine tuned a bias matrix on code snippet and text pairs. Feels more like a light version of transfer learning to me.
The article was a little unclear in the actual approach for V2 so if I have anything wrong please correct me.