54,302 blogs · 4,481,250 posts Claim your blog

Record · corpus.blog/posts/01a087e2-c24a-73e2-9f2f-861577e3a550

modded-nanogpt world record: Decoupling embedding size from model dimension

snimu.github.io · published 9 October 2025

A record is what was published and who pointed at it. The text of the post is not held here: read it at the source, or at its Wayback capture.

Read the post at snimu.github.io Wayback capture

0blogs citing

First cited: —
Most recent: —
Rank this month: not ranked
Links from this post: 10
Words captured: 20,391

Cited by

No blog we hold has cited this post yet.

Links from this post

Link

Host

https://github.com/KellerJordan/modded-nanogpt

github.com

https://github.com/KellerJordan/modded-nanogpt/pull/124

github.com

https://app.primeintellect.ai/

app.primeintellect.ai

https://x.com/johannes_hage

x.com

modded-nanogpt: Analyzing value-embedding-, UNet-, and x0-lambdas Read at the source ↗

snimu.github.io

modded-nanogpt medium world record: adding value embeddings Read at the source ↗

snimu.github.io

https://www.lesswrong.com/posts/AcKRB8wDpdaN6v6ru/interpreting-gpt-the-logit-lens

lesswrong.com

https://github.com/snimu/modded-nanogpt-experiments/blob/main/experiments/00004-x0/predictions.md

github.com

https://github.com/snimu/modded-nanogpt-experiments/blob/main/experiments/00004-x0/runs/4-2025-08-22-x00-x01-with-word-logging.py

github.com

https://github.com/ClassicLarry

github.com