R1 Merged Suite
Collection
This Deepseek "Minny" was a somewhat naive attempt at pushing the capabilities of a 1.5B model, without retraining. • 5 items • Updated
Llama-3.1-8B-Breadcrumbs-v2
This model is a 5-model integration utilizing the Model Breadcrumbs algorithm.
It merges meta-llama/Llama-3.1-8B-Instruct, Magpie-Align/Llama-3.1-8B-Magpie-Align-v0.2, allenai/Llama-3.1-Tulu-3.1-8B, and deepseek-ai/DeepSeek-R1-Distill-Llama-8B.
Hyperparameters
- Alpha ($lpha$): 0.3 (Global Scaling)
- Beta ($eta$): 0.9 (Drops bottom 90% of magnitude updates)
- Gamma ($\gamma$): 0.99 (Drops top 1% of magnitude updates)
- Downloads last month
- 2
Model tree for ForSureTesterSim/Llama-3.1-8B-R1-v2-Breadcrumbs
Base model
meta-llama/Llama-3.1-8B