@QueryloopAI goes one step further. You can upload your own set of Q&A (with or without data), and Queryloop.ai automatically evaluates and compares all major open- and closed-source LLMs in terms of accuracy, cost, and latency.
AlphaMonarch-laser performs better than base AlphaMonarch on Nous and Openllm benchmarks though it has only been trained on 50% of LoRA projections. Thanks to @erhartford, @DavidGFar, and @FernandoNetoAi for LaserQlora. Thanks to @queryloopAI for the compute.
I used @maximelabonne's NeuralMoarch-7B and performed LaserRMT scanning. Took top 16 projections with the highest SNR ratios and performed DPO on @argilla_io's Openhermes-dpo-binarized-alpha. Same hyperparams were chosen as that of AlphaMoarch training.
⚡️ Releasing AlphaMonarch-laser AlphaMonarch-laser is No. 1 on the YALL leaderboard and No. 10 on the Openllm leaderboard (7B). It has been trained with a new technique called LaserQlora. Model: huggingface.co/abideen/AlphaM…
United States 趨勢
- 1. Hato 13.4K posts
- 2. Trench 5,341 posts
- 3. New Yorkers 228K posts
- 4. Lina Khan 3,284 posts
- 5. #questpit 24.9K posts
- 6. Estevao 13.1K posts
- 7. Godzilla 20.5K posts
- 8. Gorsuch 5,618 posts
- 9. Supreme Court 133K posts
- 10. Neal Katyal 2,636 posts
- 11. Lavia 6,782 posts
- 12. IEEPA 3,775 posts
- 13. Van Jones 10.5K posts
- 14. Alastor 57.2K posts
- 15. Death Grips 2,067 posts
- 16. Qarabag 26.8K posts
- 17. Blizzcon 1,319 posts
- 18. Jacob Frey 4,669 posts
- 19. 5th of November 25.6K posts
- 20. Sauer 5,741 posts
Something went wrong.
Something went wrong.