← Back to feed

I trained a 160M code model on one GPU and got 22.56% on HumanEval

AI · Technology · 2026-10-23 21:36

The EUGEN 160M code model was trained from scratch on a single RTX PRO 6000 GPU, achieving a 22.56% pass rate on HumanEval. It supports code completion and Python-to-C++ translation while offering fast CPU inference speeds.

Source: jakubcwirlej.com


Auto-generated summary for reference only. RSS