ssd-dspark-qwen3-30b-a3b is an open-weight model from Xinze Feng. Its published files total 40.8 GB.
DSpark-family drafter (dspark), block size 7, 5 draft layers, target tap layers [1, 12, 23, 34, 45].
Model Card
DSpark-family drafter (dspark), block size 7, 5 draft layers, target tap layers [1, 12, 23, 34, 45]. Trained on the open-perfectblend corpus regenerated by Qwen3-30B-A3B (1,339,038 cache-valid rows) with a cosine LR schedule spanning 10 epochs (maxtrainsteps 26150, 2615 steps per epoch). Checkpoints at epoch boundaries live under epochN/step{N2615}/; each sits at that step of the 10-epoch schedule. Any epochN/stepS/config.json + model.safetensors pair loads like the DSpark paper's released drafters. Serve together with the target model Qwen/Qwen3-30B-A3B. 1. Rebuild the activation cache from the dataset repo xf15/ssd-perfectblend-qwen3-30b-a3b-regen with the DSpark cache builder…
Excerpt from the card by Xinze Feng.
Identity and Version
- Repository
- xf15/ssd-dspark-qwen3-30b-a3b
- Publisher
- Xinze Feng
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- 06fd5f8b9b465596cb25fec0f594cd8247daa5af
- First published
- 2026-10-01
- Last updated
- 2026-10-01
Files and Weights
29 files, 40.8 GB in total. The weights are 10 files totalling 40.7 GB in pt, safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| epoch_1/step_2615/model.safetensors | Weights | 2.0 GB | 0b56f1e04045 |
| epoch_1/step_2615/training_state.rank0.pt | Weights | 4.6 GB | fd94f56ef03c |
| epoch_1/step_2615/training_state.rank1.pt | Weights | 4.6 GB | 624b59f9bc09 |
| epoch_1/step_2615/training_state.rank2.pt | Weights | 4.6 GB | 533c6598e278 |
| epoch_1/step_2615/training_state.rank3.pt | Weights | 4.6 GB | bd37de262eb3 |
| epoch_2/step_5230/model.safetensors | Weights | 2.0 GB | 97d8521172e1 |
| epoch_2/step_5230/training_state.rank0.pt | Weights | 4.6 GB | 8a48bf19ef2f |
| epoch_2/step_5230/training_state.rank1.pt | Weights | 4.6 GB | ebb690a81643 |
| epoch_2/step_5230/training_state.rank2.pt | Weights | 4.6 GB | dbd894cbb074 |
| epoch_2/step_5230/training_state.rank3.pt | Weights | 4.6 GB | e012331645bb |
| epoch_1/step_2615/config.json | Configuration | 1.5 KB | — |
| epoch_1/step_2615/train_config.py | Configuration | 1.9 KB | — |
| epoch_2/step_5230/config.json | Configuration | 1.5 KB | — |
| epoch_2/step_5230/train_config.py | Configuration | 1.9 KB | — |
| shuffle_records/meta.json | Configuration | 255 B | — |
| shuffle_records/source_files.json | Configuration | 162 B | — |
| README.md | Documentation | 1.6 KB | — |
| shuffle_records/epoch_0.npy | Other | 10.7 MB | 6f5c9985a0ee |
| shuffle_records/epoch_1.npy | Other | 10.7 MB | dad6fe18cb65 |
| shuffle_records/epoch_2.npy | Other | 10.7 MB | 2f3b8c0302e8 |
| shuffle_records/epoch_3.npy | Other | 10.7 MB | 5cd002685e87 |
| shuffle_records/epoch_4.npy | Other | 10.7 MB | 01f3acc2b479 |
| shuffle_records/epoch_5.npy | Other | 10.7 MB | f20ac3e6b9a1 |
| shuffle_records/epoch_6.npy | Other | 10.7 MB | 133bf742df22 |
| shuffle_records/epoch_7.npy | Other | 10.7 MB | 12fd75bc13d0 |
| shuffle_records/epoch_8.npy | Other | 10.7 MB | d42955bb7ef3 |
| shuffle_records/epoch_9.npy | Other | 10.7 MB | bd0a82f0f1ad |
| shuffle_records/source_rows.npy | Other | 10.7 MB | e2d1db4147ee |
| .gitattributes | Repository | 1.5 KB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 40.7 GB
Released by Xinze Feng through its official repository on Hugging Face.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 40.7 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.