PINT Stage A: weights, k-means codebooks and model card

NOTICE

@@ -0,0 +1,6 @@
+This model is derived from facebook/hubert-base-ls960
+(https://huggingface.co/facebook/hubert-base-ls960), Copyright (c) Meta Platforms, Inc.,
+licensed under the Apache License, Version 2.0
+(https://www.apache.org/licenses/LICENSE-2.0). The encoder weights were modified by
+fine-tuning. The fine-tuned weights and every other file in this repository are released
+under CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/).

README.md

@@ -0,0 +1,58 @@
+---
+license: cc-by-nc-4.0
+library_name: pint-infer
+pipeline_tag: feature-extraction
+base_model: facebook/hubert-base-ls960
+tags:
+  - speech
+  - audio
+  - tokenizer
+  - hubert
+  - speech-tokens
+language:
+  - en
+---
+
+# PINT Stage A
+
+A HuBERT-base encoder fine-tuned on parallel utterances (the same words from different speakers and recording conditions) so that they map to the same frame-level representation. 16 kHz audio in, one 768-d vector per 20 ms frame out; token ids come from k-means codebooks.
+
+## Usage
+
+```bash
+pip install git+https://github.com/nyralabs/pint-infer
+```
+
+```python
+from pint_infer import PINTTokenizer
+
+tok = PINTTokenizer.from_pretrained("nyralabs/pint-stage-a")
+ids = tok.encode("speech.wav", method="kmeans_200")  # one token id per 20 ms
+frames = tok.embed("speech.wav")  # (T, 768) float32
+```
+
+Methods: `kmeans_50`, `kmeans_100`, `kmeans_200`, `kmeans_250`, `kmeans_500`, `kmeans_1000`.
+
+## Files
+
+| File | Content |
+|---|---|
+| `model.safetensors`, `config.json`, `preprocessor_config.json` | HuBERT-base encoder and feature extractor |
+| `kmeans_<k>.safetensors` | `(k, 768)` k-means cluster centers; assignment is nearest centroid |
+| `pint_config.json` | frame rate, codebook list |
+
+k=200 matches the vocabulary of the Stage B token head; the other codebook sizes are provided for convenience.
+
+## Intended use and limitations
+
+English speech. Input is converted to 16 kHz mono. Research and non-commercial use.
+
+## Citation
+
+A citation will be added on publication.
+
+## License
+
+The weights are released under CC BY-NC 4.0. They are derived from
+[facebook/hubert-base-ls960](https://huggingface.co/facebook/hubert-base-ls960), licensed under
+the Apache License 2.0; see `NOTICE`.

config.json

@@ -0,0 +1,76 @@
+{
+  "activation_dropout": 0.1,
+  "apply_spec_augment": true,
+  "architectures": [
+    "HubertModel"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "classifier_proj_size": 256,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_pos_batch_norm": false,
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "do_stable_layer_norm": false,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "group",
+  "feat_proj_dropout": 0.1,
+  "feat_proj_layer_norm": true,
+  "final_dropout": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.1,
+  "mask_feature_length": 10,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.05,
+  "model_type": "hubert",
+  "num_attention_heads": 12,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "transformers_version": "4.57.6",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 32
+}

kmeans_100.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1b90570ba88813a0193df7465a15ddfef08543e631236afd3b0b43b64cc6c24
+size 307288

kmeans_1000.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0aac07a05bb5cd2074378d308ca4284cfe181f1c2dd652a36f6559f79567d5e8
+size 3072096

kmeans_200.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf6202e17618784b42d55566b8ce23cbf2c47549995b5e55fb4626613056ef37
+size 614488

kmeans_250.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a28bbb8050b4bd60a3cec9091f9508830420e1bc918e675438102acf3d90bb01
+size 768088

kmeans_50.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7c90180b11bc1a4b5b4d1ce8fe7129593b2c0f53fb448fb3a0c25e3d1611dfd
+size 153688

kmeans_500.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e19582d4c1c40e476a9fd4c405c9e8c923b90536f99a508df2ef8afc72560a1
+size 1536088

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f9940a03529bcdb29b846fd8afb34c5f5bb79ad3548895b138715b5554570be
+size 377510632

pint_config.json

@@ -0,0 +1,15 @@
+{
+  "sample_rate": 16000,
+  "frame_stride_ms": 20,
+  "hidden_size": 768,
+  "base_model": "facebook/hubert-base-ls960",
+  "idseq": null,
+  "kmeans": {
+    "50": "kmeans_50.safetensors",
+    "100": "kmeans_100.safetensors",
+    "200": "kmeans_200.safetensors",
+    "250": "kmeans_250.safetensors",
+    "500": "kmeans_500.safetensors",
+    "1000": "kmeans_1000.safetensors"
+  }
+}

preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0,
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}