bf16 weights: 50% smaller, verified parity (F1/grid within noise)

README.md

@@ -1,48 +1,16 @@
 ---
 license: mit
-library_name: softchart
 tags:
-- taiko
-- rhythm-game
+- audio-to-symbolic
 - chart-generation
+- model_hub_mixin
 - music
-- audio-to-symbolic
-pipeline_tag: audio-to-audio
+- pytorch_model_hub_mixin
+- rhythm-game
+- taiko
 ---
 
-# SoftChart Generator
-
-An 8.1M-parameter encoder–decoder Transformer that generates **Taiko no Tatsujin**
-charts from music. Given a log-mel spectrogram and conditions (difficulty, density,
-and an optional song-level plan), it autoregressively produces a sequence of timed
-chart events. Part of the [SoftChart](https://github.com/JacobLinCool/SoftChart) system.
-
-This checkpoint (`pr12`) supports: **course/level/density conditioning, prev-window
-context, easy⊂hard sibling hints, style codes, and song-level plan control**
-(density envelope, breathing gaps, climax placement). Perfect difficulty
-monotonicity (Spearman 1.0) and controllable "same song → multiple valid charts".
-
-## Usage
-
-```python
-from softchart.hf import SoftChartGenerator
-from softchart.generate import load_hf, generate_song
-
-model = load_hf("JacobLinCool/softchart-generator", device="cuda")
-# mel: (128, T) log-mel at 22050 Hz, hop 256
-chart = generate_song(model, mel, course="oni", level=9, density_bucket=7, greedy=True)
-# chart["hits"] = [{"t": seconds, "type": "don"/"ka"/...}, ...]
-```
-
-Full audio→chart→TJA pipeline: [`scripts/infer.py`](https://github.com/JacobLinCool/SoftChart)
-or the [demo Space](https://huggingface.co/spaces/JacobLinCool/softchart).
-
-## Companion models
-
-- [softchart-planner](https://huggingface.co/JacobLinCool/softchart-planner) — audio → song-level plan
-- [softchart-beat](https://huggingface.co/JacobLinCool/softchart-beat) — beat/downbeat for barline anchoring
-
-## License
-
-Code and weights are **MIT**. Trained on community-made charts
-([taiko-1000-parsed](https://huggingface.co/datasets/JacobLinCool/taiko-1000-parsed)).
+This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
+- Code: https://github.com/JacobLinCool/SoftChart
+- Paper: [More Information Needed]
+- Docs: [More Information Needed]

config.json

@@ -15,10 +15,12 @@
   },
   "d_model": 256,
   "dec_layers": 4,
+  "emb_factor": null,
   "enc_layers": 4,
   "ffn": 1024,
   "func_time": false,
   "global_ctx": false,
+  "in_ch": null,
   "ptr": false,
   "vocab_size": 1802
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3829ea5f2e025c2f6cd59f55359c6fb44cff5c49c42ecf017d2e677626829f17
-size 32537124
+oid sha256:7b76b787e1f61201218dca06f2fa65f0faff33239e85f8eb7cd51f68ddb5b874
+size 16275522