Gemma 4 Training Results Lag Behind Qwen
ivan_bezdomny · x · 2026-07-17
The author mentions that they also trained a Gemma 4 model, but its current performance isn't as good as Qwen.
They observed that one characteristic of Gemma is its tendency to generate very short responses. This has an upside—it's less prone to info-dumping like GPT-5—but the downside is that the outputs are too short for continuous learning, making training difficult. The next step will likely involve adjusting the GRPO reward.
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
- 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
- OpenAI rated Astra 'Critical' for cyber capabilities — and admits it's harder to monitor — theguywhobuilds · 2026-09-11