Dev Trains 117M Custom Silia Model on H100
SrijSriv211 · reddit · 2026-07-07
A developer trained a custom 117M-parameter Silia architecture model on an H100 in 5 hours. This project is a practical follow-up to their previously published paper and was made possible by compute sponsored by community members. The model weights have been open-sourced on HuggingFace (Strawberry-s1) and GitHub, alongside a uv-based inference example. However, based on the sample outputs, the small model's generations are largely chaotic and nonsensical.
Related event: Developer Trains and Open-Sources 117M Parameter Silia Model in 5 Hours(2 posts)→
More from Research
- GigaChat Audio targets long-form audio grounding with timestamps across 120-minute inputs — ai-sage · 2026-07-21
- Paper models Transformer components as stochastic geometry and tests five architectures — Zhihua Liang · 2026-07-21
- LTX 2.3 LoRA demo changes a video’s camera angle — CQDSN · 2026-07-21
- OpenForecaster uses daily news to improve language-model forecasting — Cohere_Labs · 2026-07-21
- Baseten study finds new facts in LLM weights are fragile unless trained from many restatements — alex_verem · 2026-07-21
- uv-scripts/ocr returns to the top of Hugging Face datasets with a JSON model picker — vanstriendaniel · 2026-07-21