Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step
• 12
FINAL_Bench
Team
community
AI & ML interests
None defined yet.
Recent Activity
View all activity
Papers
View all Papers Articles
SeaWolf-AI
updated a
model 11 minutes ago
SeaWolf-AI
published a
model 11 minutes ago
SeaWolf-AI
updated a
dataset about 9 hours ago
Fantastic Model
🔥 1
10
#5 opened 2 days ago
by 
apollo-mg
SeaWolf-AI
updated a
collection about 13 hours ago
SeaWolf-AI
updated a
Space about 13 hours ago
SeaWolf-AI
published a
Space about 13 hours ago
SeaWolf-AI
updated a
collection 1 day ago
Add paper link, author information, and citation
#1 opened 1 day ago
by 
nielsr
Improve model card metadata and link to paper
🔥 1
#3 opened 1 day ago
by nielsr
Added vision: Qwen3.6-35B-A3B-DarleyQuinn
👍 1
18
#4 opened 4 days ago
by 
nightmedia
APEX Quant Request + Real World Performance
6
#3 opened 18 days ago
by 
el4
SeaWolf-AI
published an article 2 days ago
Article
Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step
FINAL-Bench
• • 12 SeaWolf-AI
posted an update 2 days ago
Post
5125
🧬 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: Darwin Family: MRI-Trust-Weighted Evolutionary Merging for Training-Free Scaling of Language-Model Reasoning (2605.14386)
🔗 arXiv: https://arxiv.org/abs/2605.14386
🔗 Model: 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 🙌
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: Darwin Family: MRI-Trust-Weighted Evolutionary Merging for Training-Free Scaling of Language-Model Reasoning (2605.14386)
🔗 arXiv: https://arxiv.org/abs/2605.14386
🔗 Model: 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 🙌
- 3 replies
seawolf2357
submitted a
paper to Daily Papers 2 days ago
SeaWolf-AI
updated a
model 2 days ago
SeaWolf-AI
updated a
Space 2 days ago
SeaWolf-AI
updated 3
models 2 days ago