nanoGPT · Issues· 349 open

test

Realtime response generation simulation instead of one big response at the end.

GPT.generate() crashes on temperature=0 instead of doing greedy decoding

Training on Intel(R) Data Center GPU Max 1550 GPUs (formerly codenamed Ponte Vecchio)

Code quality scan: 9 findings (C-, 68/100)

Is this loss curve normal

Pretraining loss explosion

Pretraining Divergence

Train/Val Loss Issues when training GPT-2 from OWT

During the training, the loss first drops and then rises

Critical Sandbox Escape Vulnerability in nanochat

the model is not domain agnostic

Qodo AntiSlop scan found 8 issues across 7 recent PRs

small warning

Scalar C++ Transformer Implementation — Verified Against PyTorch