Fine-tuning Gemma Embedding 2 on 3,000 hours of Vietnamese audio hits SOTA for ~$10
MaziyarPanahi · x · 2026-10-14
A developer fine-tuned Google's Gemma Embedding 2 on 3,000 hours of Vietnamese audio and reached SOTA, training just 2 hours on 8xH100 — roughly $10 of compute by one estimate — showing how cheap language-specific specialization has become.
Related event: Gemma Embedding 2 Fine-tuned to SOTA on Vietnamese Speech in 2 Hours(2 posts)→
More from Models
- Dev behind $1.27M AI DM operation details prompt drift and silent API degradation — alanshore222 · 2026-10-14
- Gemini's forced "Automatic" model selection leaves users calling it dumb as bread — InsideOil3078 · 2026-10-14
- Gemini 4 Argon greets evaluators inside its chain of thought, Apollo Research finds — daemon86 · 2026-10-14
- Open-Weight Models Go "Abliterated" Within 24 Hours, Powering $14.88 Extremist Chatbots — nordicinst · 2026-10-14
- Frontier models get stale in two weeks, user says native video and audio are the missing piece — imjustnewatai · 2026-10-14
- Microsoft Patches 1,400+ Flaws in One Month, Credits LLMs for Bug Discovery — suchenzang · 2026-10-14