Clarify Audio-JEPA training differences from the ICME checkpoint
Browse files- README.md +12 -2
- export_manifest.json +1 -1
README.md
CHANGED
|
@@ -24,8 +24,16 @@ patches from an exponential-moving-average teacher. This export contains the
|
|
| 24 |
**student encoder**, matching the research model's inference path; the pretraining
|
| 25 |
predictor and teacher are omitted.
|
| 26 |
|
| 27 |
-
This
|
| 28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
|
| 30 |
This repository contains the trained encoder, its preprocessing configuration,
|
| 31 |
and the custom Transformers implementation. No installation of the research
|
|
@@ -55,6 +63,8 @@ weight to the Speech, Music, and Environment category means.
|
|
| 55 |
|
| 56 |
Audio-JEPA scores were supplied by the author for [run `jp6l70l6`](https://wandb.ai/tuncay-ludovic/audio%20embeddings/runs/jp6l70l6). The remaining scores are reported in the [project README](https://github.com/LudovicTuncay/audio-embeddings/blob/bb88bf790b1dcf8251c6b38e7a4766534adf33d3/README.md). These are reported research results, not a new benchmark run of the Transformers exports.
|
| 57 |
|
|
|
|
|
|
|
| 58 |
## Model and training
|
| 59 |
|
| 60 |
| Property | Value |
|
|
|
|
| 24 |
**student encoder**, matching the research model's inference path; the pretraining
|
| 25 |
predictor and teacher are omitted.
|
| 26 |
|
| 27 |
+
**This is an updated training run, not an exact reproduction of the ICME 2025 model.**
|
| 28 |
+
|
| 29 |
+
| Version | Audio sample rate | Training steps |
|
| 30 |
+
| --- | --- | --- |
|
| 31 |
+
| Original ICME 2025 model ([ltuncay/Audio-JEPA](https://huggingface.co/ltuncay/Audio-JEPA)) | 32 kHz | 100,000 |
|
| 32 |
+
| This release, run `jp6l70l6` | 16 kHz | 200,000 |
|
| 33 |
+
|
| 34 |
+
The author reports better X-ARES results for this newer checkpoint. The Audio-JEPA
|
| 35 |
+
scores below are for this release and should not be presented as the original
|
| 36 |
+
ICME results. The original checkpoint remains available in its own repository.
|
| 37 |
|
| 38 |
This repository contains the trained encoder, its preprocessing configuration,
|
| 39 |
and the custom Transformers implementation. No installation of the research
|
|
|
|
| 63 |
|
| 64 |
Audio-JEPA scores were supplied by the author for [run `jp6l70l6`](https://wandb.ai/tuncay-ludovic/audio%20embeddings/runs/jp6l70l6). The remaining scores are reported in the [project README](https://github.com/LudovicTuncay/audio-embeddings/blob/bb88bf790b1dcf8251c6b38e7a4766534adf33d3/README.md). These are reported research results, not a new benchmark run of the Transformers exports.
|
| 65 |
|
| 66 |
+
The Audio-JEPA row refers to the newer **16 kHz, 200,000-step** run, not the original ICME model (**32 kHz, 100,000 steps**). The author reports better results for this newer checkpoint.
|
| 67 |
+
|
| 68 |
## Model and training
|
| 69 |
|
| 70 |
| Property | Value |
|
export_manifest.json
CHANGED
|
@@ -31,7 +31,7 @@
|
|
| 31 |
},
|
| 32 |
"files": {
|
| 33 |
"CODE_LICENSE": "8fe9e8b749cd4abedabcb3100df445db899e72192394d14cc2dbf24a40811af6",
|
| 34 |
-
"README.md": "
|
| 35 |
"adapters.py": "ca0f26826763c0e48b7508da732242bb83adfeb4814e389ed3fde93ad648c728",
|
| 36 |
"config.json": "e98990569657da3b4b3046b3eef296ebdf0807aa00394c77e4f4f8228e60a638",
|
| 37 |
"configuration_audio.py": "59f3a0b8db0df5af85e677ac33a4431595e9b2eff6d34c1e6a1778dd75a4568d",
|
|
|
|
| 31 |
},
|
| 32 |
"files": {
|
| 33 |
"CODE_LICENSE": "8fe9e8b749cd4abedabcb3100df445db899e72192394d14cc2dbf24a40811af6",
|
| 34 |
+
"README.md": "d4237d760b2c8a07d7e2a8711d0c74312118d0e566575204376753fcb235cb78",
|
| 35 |
"adapters.py": "ca0f26826763c0e48b7508da732242bb83adfeb4814e389ed3fde93ad648c728",
|
| 36 |
"config.json": "e98990569657da3b4b3046b3eef296ebdf0807aa00394c77e4f4f8228e60a638",
|
| 37 |
"configuration_audio.py": "59f3a0b8db0df5af85e677ac33a4431595e9b2eff6d34c1e6a1778dd75a4568d",
|