mineself2016 commited on
Commit
3d0c815
·
verified ·
1 Parent(s): 6a44f5b

Sync latest GeneMamba docs and next-token training updates

Browse files
Files changed (1) hide show
  1. README.md +12 -45
README.md CHANGED
@@ -26,7 +26,6 @@ A Hugging Face compatible implementation of GeneMamba, a foundational state-spac
26
  - [Datasets](#datasets)
27
  - [Usage Guide](#usage-guide)
28
  - [Citation](#citation)
29
- - [License](#license)
30
 
31
  ---
32
 
@@ -316,14 +315,14 @@ Task-Specific Heads:
316
 
317
  ### Key Design Choices
318
 
319
- - **Sequence Processing**: Bidirectional Mamba with multiple aggregation modes (mean/sum/concat/gate)
320
- - **Pooling Strategy**: Mean pooling over sequence (CLS token available as option)
321
  - **Regularization**: Dropout on classification head
322
  - **Activation**: No explicit activation (Mamba uses internal gating)
323
 
324
  ---
325
 
326
- ## Usage Guide
327
 
328
  ### Loading Models
329
 
@@ -387,7 +386,7 @@ config = model.config
387
  config.hidden_dropout_prob = 0.2
388
  ```
389
 
390
- ---
391
 
392
  ## Important Notes ⚠️
393
 
@@ -436,26 +435,20 @@ See the `examples/` directory for complete scripts:
436
 
437
  - `1_extract_embeddings.py` - Extract cell embeddings
438
  - `2_finetune_classification.py` - Cell type annotation
439
- - `4_pretrain_from_scratch.py` - Train from scratch (next-token + optional resume)
440
-
441
- Run any example:
442
-
443
- ```bash
444
- python examples/01_extract_embeddings.py
445
- ```
446
 
447
  ---
448
 
449
  ## Citation
450
 
451
- If you use GeneMamba in your research, please cite:
452
 
453
  ```bibtex
454
- @article{genemamba2024,
455
- title={GeneMamba: A Foundation Model for Single-Cell Analysis},
456
- author={Contributors...},
457
- journal={bioRxiv},
458
- year={2024}
459
  }
460
  ```
461
 
@@ -515,30 +508,4 @@ GeneMamba_repo/
515
 
516
  ---
517
 
518
- ## Contributing
519
-
520
- Contributions welcome! Please:
521
-
522
- 1. Fork the repository
523
- 2. Create a feature branch
524
- 3. Submit a pull request
525
-
526
- ---
527
-
528
- ## License
529
-
530
- This project is licensed under the Apache 2.0 License - see [LICENSE](LICENSE) for details.
531
-
532
- ---
533
-
534
- ## Support
535
-
536
- - 📖 **Documentation**: See `docs/` directory
537
- - 🐛 **Issues**: Report on GitHub
538
- - 💬 **Discussions**: Join our community forum
539
- - 📧 **Email**: Support contact (to be added)
540
-
541
- ---
542
-
543
- **Last Updated**: March 2026
544
- **Maintained by**: GeneMamba Team
 
26
  - [Datasets](#datasets)
27
  - [Usage Guide](#usage-guide)
28
  - [Citation](#citation)
 
29
 
30
  ---
31
 
 
315
 
316
  ### Key Design Choices
317
 
318
+ - **Bidirectional Mamba Block**: Bidirectional Mamba enables significant improvement in gene rank reconstruction task
319
+ - **Pooling Strategy**: Bidirectional Mamba with multiple aggregation modes (mean/sum/concat/gate)
320
  - **Regularization**: Dropout on classification head
321
  - **Activation**: No explicit activation (Mamba uses internal gating)
322
 
323
  ---
324
 
325
+ <!-- ## Usage Guide
326
 
327
  ### Loading Models
328
 
 
386
  config.hidden_dropout_prob = 0.2
387
  ```
388
 
389
+ --- -->
390
 
391
  ## Important Notes ⚠️
392
 
 
435
 
436
  - `1_extract_embeddings.py` - Extract cell embeddings
437
  - `2_finetune_classification.py` - Cell type annotation
438
+ - `3_pretrain_from_scratch.py` - Train from scratch (next-token + optional resume)
 
 
 
 
 
 
439
 
440
  ---
441
 
442
  ## Citation
443
 
444
+ If you find GeneMamba is useful in your research, please cite:
445
 
446
  ```bibtex
447
+ @article{qi2025genemamba,
448
+ title={GeneMamba: An Efficient and Effective Foundation Model on Single Cell Data},
449
+ author={Qi, Cong and Fang, Hanzhang and Jiang, Siqi and Song, Xun and Hu, Tianxing and Zhi, Wei},
450
+ journal={arXiv preprint arXiv:2504.16956},
451
+ year={2026}
452
  }
453
  ```
454
 
 
508
 
509
  ---
510
 
511
+ **Last Updated**: March 2026