TRAINING
1 item
1 post
Blog
Distilling an LLM on One GPU: Offline Top-K Logits and a Fused Chunked KL LossLLM distillation on one GPU: cache the teacher's top-100 logits offline and use a fused chunked KL loss to cut a GPT-OSS 20B run from four nodes to one.
Get Smarter About AI Dev
New tutorials, open-source projects, and deep dives on coding agents - delivered weekly.
One email per weekReal code, not theoryFree forever