Views
No views yet
| Seed | AUC | MRR | NDCG@5 | NDCG@10 |
|---|---|---|---|---|
| 42 * | 0.6614 | 0.3135 | 0.3469 | 0.4096 |
1model:
2 name: nrms
3 architecture:
4 news_encoder:
5 type: multi_head_self_attention
6 num_heads: 12
7 head_dim: 15
8 attention_hidden_dim: 200
9 user_encoder:
10 type: multi_head_self_attention
11 num_heads: 20
12 head_dim: 15
13 attention_hidden_dim: 200
14 click_predictor:
15 type: dot_product
16 text_pooler:
17 type: attention
18 num_heads: 6
19 head_dim: 128
20 attention_query_dim: 200
21 dropout_rate: 0.0
22 embedding:
23 size: 300
24 trainable: true
25 dropout_rate: 0.2
26 seed: 42
27inputs:
28 title:
29 max_length: 32
30 history:
31 max_length: 50
32 impressions:
33 max_length: 5
34 process_title: true
35 process_abstract: false
36 process_category: false
37 process_subcategory: false
38 process_user_id: false
39training:
40 loss:
41 name: categorical_crossentropy
42 from_logits: true
43 reduction: sum_over_batch_size
44 label_smoothing: 0.0
45 optimizer: adam
46 learning_rate: 0.0001
47 batch_size: 256
48 num_epochs: 10
49 early_stopping:
50 patience: 3
51 min_improvement: 0.01
52 negative_sampling:
53 strategy: random
54 candidates: 4
55evaluation:
56 mode: fast
57 metrics:
58 - auc
59 - mrr
60 - ndcg@5
61 - ndcg@10
62 batch_size: 256newsrex/NRMS-JAX-MIND-small-bert-random/
├── model.safetensors ← best seed (42)
├── test_results.json
├── training_run_summary.json
├── seed_42/model.safetensors
└── README.md1git clone https://github.com/igor17400/NewsReX.git
2cd NewsReX && uv sync
3
4# Run evaluation with best seed weights
5uv run python src/eval.py \
6 experiment=mind/glove/nrms \
7 framework=jax \
8 weights=hf://newsrex/NRMS-JAX-MIND-small-bert-random/model.safetensors
9
10# Run evaluation with a specific seed
11uv run python src/eval.py \
12 experiment=mind/glove/nrms \
13 framework=jax \
14 weights=hf://newsrex/NRMS-JAX-MIND-small-bert-random/seed_42/model.safetensors1@misc{newsrex2026,
2 title={NewsReX: An Open-Source Multi-Framework for Neural News Recommendation},
3 author={Igor L. R. Azevedo and Toyotaro Suzumura and Yuichiro Yasui},
4 year={2025},
5 eprint={2508.21572},
6 archivePrefix={arXiv},
7 primaryClass={cs.IR},
8 url={https://arxiv.org/abs/2508.21572},
9}