Controversy Over Nemotron 3 Nano Reproduction
JJitsev · x · 2026-07-13
The post discusses the reliability of the reproduction evaluation for Nemotron 3 Nano.
- The author points out a clear discrepancy between NVIDIA's original report and soofi's report; for metrics like HumanEval, GSM8K, and MMLU-Pro, Nemotron 3 Nano performs worse in soofi's evaluation.
- Because these numbers are not aligned, it's unclear if a valid reproduction of Nemotron 3 Nano has actually been achieved.
- A reply mentions that soofi's scores are slightly higher than Nemotron 3 Nano, looking like a successful reproduction on the surface; but if the original Nemotron 3 Nano evaluation was flawed, the comparison becomes uncertain.
- The main takeaway: once benchmark numbers are inconsistent, conclusions about model reproduction lose credibility.
Related event: Controversy Over Nemotron Reproduction and Custom Benchmarks(4 posts)→
More from Models
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- 105 hidden bugs, 2 repos: DeepSeek V4.1 Flash fixes 24 at $1.80 vs Opus 5's 27 at $51.33 — ChartsJournalX · 2026-09-11
- awesome-llm-leaderboards: an open-source directory of LLM leaderboards, pricing tables, comparison tools — Last_Establishment_1 · 2026-09-11
- Anthropic claims it works to keep eval environments unidentifiable to models — MaxKannen · 2026-09-11
- Nex N2.5 Pro released on Hugging Face with 407GB of weights — jinnyjuice · 2026-09-11
- RoMa v2 image matching model unveiled in the usual black poster — ducha_aiki · 2026-09-11