Finetuning Sessions – Week 4 Office Hours

Here's the recording from yesterday's live session, covering the Week 4 content of the Finetuning Sessions.

During office hours, we focused on:

• The VRAM bottleneck in LLM training and why memory — not compute — is the real constraint

• How number formats evolved from FP32 to 4-bit, and why NF4 is purpose-built for neural network weights

• The difference between