Hugging Face's logo
beans of sloar's picture
11 262

beans of sloar

solarbeams
ausntmarzi's profile picture honey90's profile picture EX4L's profile picture
·

AI & ML interests

None yet

Recent Activity

liked a model 4 days ago
FINAL-Bench/Darwin-28B-REASON
liked a Space 4 days ago
FINAL-Bench/Darwin-9B-NEG
reacted to SeaWolf-AI's post with 🤗 6 days ago
🧬 Darwin Family: Zero Gradient Steps, GPQA Diamond 88.89% How far can we push LLM reasoning *without* training? Our team at VIDRAFT submitted this paper to Daily Papers yesterday, and it's currently #3. Huge thanks to everyone who upvoted — sharing the core ideas below. 🔗 Paper: https://huggingface.co/papers/2605.14386 🔗 arXiv: https://arxiv.org/abs/2605.14386 🔗 Model: https://huggingface.co/FINAL-Bench/Darwin-28B-REASON 🔗 Model: https://huggingface.co/FINAL-Bench/Darwin-28B-Opus --- TL;DR Darwin Family is a training-free evolutionary merging framework. By recombining the weight spaces of existing LLM checkpoints — with zero gradient-based training — it reaches frontier-level reasoning. - 🏆 Darwin-28B-Opus: GPQA Diamond 88.89% - 💸 Zero gradient steps — not a single B200 or H200 hour needed - 🧬 Consistent gains across 4B → 35B scale - 🔀 Cross-architecture breeding between Transformer and Mamba families - 🔁 Stable recursive multi-generation evolution #Three Core Mechanisms ① 14-dim Adaptive Merge Genome — fine-grained recombination at both component level (Attention / FFN / MLP / LayerNorm / Embedding) and block level, expanding the prior evolutionary-merge search space. ② MRI-Trust Fusion — we diagnose each layer's reasoning contribution via an **MRI (Model Reasoning Importance)** signal and fuse it with evolutionary search through a **learnable trust parameter**. Trust the diagnostic too much and search collapses; ignore it and search becomes inefficient — Darwin learns the balance from data. ③ Architecture Mapper — weight-space breeding across heterogeneous families. Attention × SSM crossover actually works. Why It Matters > Diagnose latent capabilities already encoded in open checkpoints, > and recombine them — no gradients required. Replies and critiques welcome 🙌
View all activity

Organizations

None yet