Sentence Transformers adds model merging: 60/40 weighted blend of two MiniLM checkpoints
tomaarsen · x · 2026-10-29
Sentence Transformers maintainer tomaarsen demonstrates the library's model merging feature, which combines compatible checkpoints without another training run. The example merges all-MiniLM-L6-v2 and multi-qa-MiniLM-L6-cos-v1 via the linear method with 0.6/0.4 weights, saves the result, and encodes text normally (output shape (2, 384)). He reports best results merging checkpoints fine-tuned from the same base model.
Related event: Sentence Transformers Adds Training-Free Model Merging(2 posts)→
More from Models
- Diffusion models at ~50% Fortune 500 adoption with very low spend, still early — deanwball · 2026-10-29
- Mistral Large 4 launch-day interview: open weights, own data centers, excels at cyber — huggingface · 2026-10-29
- Gamma 5 takes #1 on Design Arena for slides, beating Claude Opus 5.5 — haroonc · 2026-10-29
- Mistral Open-Sources Trillion-Parameter Coding Model Le Chonk, Still Trails DeepSeek and GLM — The Decoder · 2026-10-29
- Mamba co-author credits SmolLM3's open recipe and checkpoints for enabling research — srush_nlp · 2026-10-29
- ThursdAI Recap: Gemini 4 Carbon, Genie 4, Anthropic IPO Rumors and More — WolframRvnwlf · 2026-10-29