REAM
Collection
Compressed MoE models with a reduced number of experts. See additional models at https://huggingface.co/bknyaz. • 11 items • Updated • 5
AI & ML interests
None defined yet.
Recent Activity
Papers
REAM: Merging Improves Pruning of Experts in LLMs
Less is More: Recursive Reasoning with Tiny Networks
bknyaz
updated a
collection 1 day ago
keep doing it!
❤️ 1
2
#4 opened 18 days ago
by 
pirola
bknyaz
updated a
collection 9 days ago
bknyaz
updated 2
models 9 days ago
bknyaz
published a
model 9 days ago
bknyaz
updated a
model 9 days ago
bknyaz
published a
model 9 days ago
bknyaz
submitted a
paper to Daily Papers 10 days ago
bknyaz
updated 2
models 10 days ago
bknyaz
authored 6
papers 10 days ago
(Almost) Free Modality Stitching of Foundation Models
Paper • 2507.10015 • Published • 2
M^4olGen: Multi-Agent, Multi-Stage Molecular Generation under Precise Multi-Property Constraints
Paper • 2601.10131 • Published • 17
Recursive Autoconvolution for Unsupervised Learning of Convolutional Neural Networks
Paper • 1606.00611 • Published
Celo2: Towards Learned Optimization Free Lunch
Paper • 2602.19142 • Published
Celo: Training Versatile Learned Optimizers on a Compute Diet
Paper • 2501.12670 • Published
REAM: Merging Improves Pruning of Experts in LLMs
Paper • 2604.04356 • Published • 8
Has REAM been checked for their resilience towards quantization?
4
#1 opened 3 months ago
by 
TomLucidor
Testing REAM on Kimi-Linear and Nemtron's hybrid attention models
2
#2 opened 2 months ago
by TomLucidor