You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

NEXUS Shared Expert Weights (10K Steps)

Trained shared expert weights from NEXUS (Neural Expert Unified Specialization) calibration run.

Model Details

Base Model: GPT-OSS 120B
Training Steps: 10,000
Method: Top-24 PCA-selected experts, frozen router
Parameters: 896,106,240 (shared expert only)
Size: 1.67GB (BF16)
Training Config: Frozen router, advanced scheduler, KL distillation

What This Contains

This file contains ONLY the shared expert weights (216 parameter tensors) from a NEXUS-trained model.

To use:

Start with base GPT-OSS 120B model
Add NEXUS shared expert architecture
Load these weights

Usage

import torch

# Load weights
shared_weights = torch.load("nexus_shared_expert_weights_10k.pt")

# Apply to model with NEXUS architecture
model.load_state_dict(shared_weights, strict=False)

About NEXUS

NEXUS enables efficient domain specialization of massive MoE models by training a small shared expert while keeping routed experts frozen.

See: https://github.com/yourusername/nexus

License

MIT

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support