Downloads · 30 days
0
week2-moe-hash-router
Deepu1965/week2-moe-hash-router
week2-moe-hash-router is a machine learning model from Deepu1965. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Best validation loss: 5.6068 Top-k: 1 Aux loss coef: 0.0
Access
Public
Updated Nov 11, 2025
Repo size
256 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt86 MB · 99%
At a glance
- Access
- Public
- Created
- Nov 2, 2025
- Updated
- Nov 11, 2025
- SHA
- 7c4b4af5
Model card
From the Hugging Face model README
Week 2 MoE Seq2Seq (hash routing)
- Best validation loss: 5.6068
- Top-k: 1
- Aux loss coef: 0.0
Artifacts include the trained state dict (model.pt), metrics (metrics.json), per-epoch history (history.csv), and tokenizer files.
Architecture
- Encoder-Decoder Transformer with Sparse MoE layers
- Hash-based routing (deterministic) or Token-choice top-k routing (learned)
- Load balancing auxiliary loss for top-k routing
- Trained from scratch on XSum for abstractive summarization
- Created
- Nov 2, 2025
- Updated
- Nov 11, 2025