Add the crosslingual release with its quantized builds

README.md

@@ -0,0 +1,182 @@
+---
+language:
+- ru
+- en
+- de
+- pl
+- it
+- fr
+- es
+- bn
+license: mit
+library_name: transformers
+pipeline_tag: audio-classification
+base_model: Aniemore/unispeech-sat-emotion-russian-resd
+base_model_relation: finetune
+datasets:
+- Aniemore/resd
+- Aniemore/resd_annotated
+- amu-cai/CAMEO
+tags:
+- audio-classification
+- emotion-recognition
+- speech-emotion-recognition
+- speech
+- multilingual
+- russian
+- quantized
+- compressed-tensors
+- int8
+- fp8
+- int4
+metrics:
+- f1
+- accuracy
+- recall
+model-index:
+- name: unispeech-sat-emotion-v1-crosslingual
+  results:
+  - task:
+      name: Speech Emotion Recognition
+      type: audio-classification
+    dataset:
+      name: RESD test
+      type: Aniemore/resd
+    metrics:
+    - name: Macro F1
+      type: f1
+      value: 0.5697
+    - name: Unweighted accuracy
+      type: recall
+      value: 0.6158
+  - task:
+      name: Speech Emotion Recognition
+      type: audio-classification
+    dataset:
+      name: Dusha podcast test
+      type: dusha
+    metrics:
+    - name: Macro F1
+      type: f1
+      value: 0.3705
+    - name: Unweighted accuracy
+      type: recall
+      value: 0.6992
+  - task:
+      name: Speech Emotion Recognition
+      type: audio-classification
+    dataset:
+      name: CAMEO test
+      type: amu-cai/CAMEO
+    metrics:
+    - name: Macro F1
+      type: f1
+      value: 0.5689
+    - name: Unweighted accuracy
+      type: recall
+      value: 0.5706
+---
+
+<img src="assets/banner.svg" alt="unispeech-sat-emotion-v1-crosslingual" width="100%">
+
+# unispeech-sat-emotion-v1-crosslingual
+
+Speech emotion recognition over seven classes &mdash; `anger`, `disgust`, `enthusiasm`, `fear`, `happiness`, `neutral`, `sadness`.
+
+Same architecture as [`Aniemore/unispeech-sat-emotion-russian-resd`](https://huggingface.co/Aniemore/unispeech-sat-emotion-russian-resd), retrained on a mix of 27,939 clips spanning eight languages and three speaking registers instead of one acted Russian corpus. The point of the change is spontaneous speech: the previous release was trained only on acted dialogue, where every class is equally frequent and every utterance is performed, and real speech is neither.
+
+Quantized builds ship in the same repository under `int8/`, `fp8/` and `int4/`.
+
+## Results
+
+| test set | what it is | macro-F1 | UA | WA | previous release |
+|---|---|---:|---:|---:|---:|
+| RESD test | acted Russian, 7 balanced classes | **0.5697** | 0.6158 | 0.6179 | 0.7027 |
+| Dusha podcast test | spontaneous Russian, majority neutral | **0.3705** | 0.6992 | 0.6993 | 0.1000 |
+| CAMEO test | 7 non-Russian languages | **0.5689** | 0.5706 | 0.6832 | 0.1946 |
+
+<img src="assets/panel.svg" alt="Panel results" width="760">
+
+Read the first two rows together. The acted score goes down and the spontaneous score goes up; both follow from the same change, and which one matters is a deployment question. If your audio is read or performed speech, the previous release may still suit you better.
+
+<details><summary>About the CAMEO row</summary>
+
+CAMEO ships no train/test partition, and the usual way to make one &mdash; a random split over clips &mdash; puts nearly every test speaker into training as well: three of its twelve constituent corpora contain a single speaker each, so no clip-level split of them can be speaker-disjoint even in principle. The number above is reported for completeness. Treat it as an in-domain figure, not as evidence of cross-lingual transfer.
+
+</details>
+
+### Per-class recall on spontaneous speech
+
+| class | this model | previous release |
+|---|---:|---:|
+| `angry` | 0.5689 | 0.5629 |
+| `neutral` | 0.6958 | 0.0583 |
+| `positive` | 0.8138 | 0.5029 |
+| `sad` | 0.7184 | 0.2718 |
+
+`neutral` carries most of real speech and is the class the previous release missed.
+
+## Quantized variants
+
+| subfolder | scheme | weights | vs fp32 | macro-F1 | UA | WA |
+|---|---|---:|---:|---:|---:|---:|
+| _(root)_ | fp32 | 1206 MiB | 1.0x | 0.5697 | 0.6158 | 0.6179 |
+| `int8` | W8A16 | 351 MiB | 3.4x smaller | 0.5697 | 0.6158 | 0.6179 |
+| `fp8` | W8A16-float | 343 MiB | 3.5x smaller | 0.5663 | 0.6126 | 0.6143 |
+| `int4` | W4A16_ASYM | 209 MiB | 5.8x smaller | 0.5714 | 0.6163 | 0.6179 |
+
+<img src="assets/quality.svg" alt="Quality after quantization" width="760">
+
+<img src="assets/size.svg" alt="Weights on disk" width="760">
+
+Weight-only, round-to-nearest, no calibration. Every variant lands within the seed spread of the fp32 parent on RESD test, so the choice is about download size rather than about quality.
+
+## Training data
+
+| corpus | clips | language | register |
+|---|---:|---|---|
+| RESD | 948 | Russian | acted dialogue |
+| Dusha crowd | 6,800 | Russian | acted, crowd-sourced |
+| CAMEO | 6,800 | 7 languages | 12 corpora, no Russian |
+| Dusha podcast | 6,060 | Russian | spontaneous podcast speech |
+| IEMOCAP | 4,735 | English | elicited dyadic sessions |
+| ASVP-ESD | 2,596 | multilingual | mixed register |
+| **total** | **27,939** | 8 languages | 3 registers |
+
+A slice is held out of every corpus in the mix, in the same proportion, and model selection is on that held-out split &mdash; never on any of the test sets above. Labels are unified to seven classes; four-class corpora are mapped upward and scored on the classes they actually contain.
+
+## Usage
+
+```python
+import torch, librosa
+from transformers import AutoModelForAudioClassification, AutoFeatureExtractor
+
+repo = "Aniemore/unispeech-sat-emotion-v1-crosslingual"
+model = AutoModelForAudioClassification.from_pretrained(repo).eval()
+fe = AutoFeatureExtractor.from_pretrained(repo)
+
+# Resample to 16 kHz. Do not skip it: RESD itself ships at 44.1 kHz,
+# and handing the model 44.1 kHz audio while telling the extractor it
+# is 16 kHz stretches time 2.8x and silently changes the answer.
+wav, _ = librosa.load("clip.wav", sr=16000, mono=True)
+x = fe(wav, sampling_rate=16000, return_tensors="pt", padding=True)
+with torch.no_grad():
+    probs = model(**x).logits.softmax(-1)[0]
+print({model.config.id2label[i]: round(p.item(), 3) for i, p in enumerate(probs)})
+```
+
+For a quantized build, name the subfolder &mdash; only that subfolder is downloaded:
+
+```python
+model = AutoModelForAudioClassification.from_pretrained(
+    repo, subfolder="int8").eval()          # or "fp8", "int4"
+fe = AutoFeatureExtractor.from_pretrained(repo, subfolder="int8")
+```
+
+## Limitations
+
+- Scores are the mean of two seeds; the seed spread on the 280-clip RESD split is ±0.03&ndash;0.05, so differences smaller than that are not differences.
+- The spontaneous set has four classes where the model has seven, so its numbers are computed over a mapped label space and are not comparable to seven-class figures.
+- Spontaneous scores are at zero decision bias. Calibrating the `neutral` threshold on your own development split will move them.
+- Inherited from [`microsoft/unispeech-sat-large`](https://huggingface.co/microsoft/unispeech-sat-large); the licence follows the base model.
+

config.json

@@ -0,0 +1,126 @@
+{
+  "activation_dropout": 0.05,
+  "apply_spec_augment": true,
+  "architectures": [
+    "UniSpeechSatForSequenceClassification"
+  ],
+  "attention_dropout": 0.05,
+  "bos_token_id": 1,
+  "classifier_proj_size": 768,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.05,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.05,
+  "finetuning_task": [
+    "unispeech_sat_classification"
+  ],
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.05,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "anger",
+    "1": "disgust",
+    "2": "enthusiasm",
+    "3": "fear",
+    "4": "happiness",
+    "5": "neutral",
+    "6": "sadness"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "anger": 0,
+    "disgust": 1,
+    "enthusiasm": 2,
+    "fear": 3,
+    "happiness": 4,
+    "neutral": 5,
+    "sadness": 6
+  },
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "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": "unispeech-sat",
+  "num_attention_heads": 16,
+  "num_clusters": 504,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "num_negatives": 100,
+  "pad_token_id": 0,
+  "pooling_mode": "mean",
+  "problem_type": "single_label_classification",
+  "proj_codevector_dim": 768,
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "transformers_version": "5.14.1",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 40,
+  "xvector_output_dim": 512
+}

fp8/config.json

@@ -0,0 +1,165 @@
+{
+  "activation_dropout": 0.05,
+  "apply_spec_augment": true,
+  "architectures": [
+    "UniSpeechSatForSequenceClassification"
+  ],
+  "attention_dropout": 0.05,
+  "bos_token_id": 1,
+  "classifier_proj_size": 768,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.05,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.05,
+  "finetuning_task": [
+    "unispeech_sat_classification"
+  ],
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.05,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "anger",
+    "1": "disgust",
+    "2": "enthusiasm",
+    "3": "fear",
+    "4": "happiness",
+    "5": "neutral",
+    "6": "sadness"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "anger": 0,
+    "disgust": 1,
+    "enthusiasm": 2,
+    "fear": 3,
+    "happiness": 4,
+    "neutral": 5,
+    "sadness": 6
+  },
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "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": "unispeech-sat",
+  "num_attention_heads": 16,
+  "num_clusters": 504,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "num_negatives": 100,
+  "pad_token_id": 0,
+  "pooling_mode": "mean",
+  "problem_type": "single_label_classification",
+  "proj_codevector_dim": 768,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "naive-quantized",
+        "input_activations": null,
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "channel",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "naive-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "unispeech_sat.feature_projection.projection",
+      "projector",
+      "classifier"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.17.1"
+  },
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "transformers_version": "5.10.1",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 40,
+  "xvector_output_dim": 512
+}

fp8/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4451d1cae24923bffa3bdd10c2d34d68fb7dbc0365eeda53e5638c6c484d74b
+size 359898548

fp8/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": true,
+  "sampling_rate": 16000
+}

fp8/recipe.yaml

@@ -0,0 +1,27 @@
+default_stage:
+  default_modifiers:
+    QuantizationModifier:
+      config_groups:
+        group_0:
+          targets: [Linear]
+          weights:
+            num_bits: 8
+            type: float
+            symmetric: true
+            group_size: null
+            strategy: channel
+            block_structure: null
+            dynamic: false
+            actorder: null
+            scale_dtype: null
+            zp_dtype: null
+            observer: memoryless_minmax
+            observer_kwargs: {}
+          input_activations: null
+          output_activations: null
+          format: null
+      targets: [Linear]
+      ignore: ['re:.*feature_extractor.*', 're:.*feature_projection.*', 're:.*pos_conv_embed.*',
+        're:.*gru_rel_pos_linear$', projector, classifier, 're:.*audio_projector$', 're:.*text_projector$',
+        're:.*fusion_module.*', 're:.*text_model\.pooler.*']
+      bypass_divisibility_checks: false

fp8/special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

fp8/tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "bos_token": "<s>",
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "Ar4ikov/unispeech-sat-projector-russian-resd",
+  "pad_token": "<pad>",
+  "processor_class": "Wav2Vec2Processor",
+  "replace_word_delimiter_char": " ",
+  "special_tokens_map_file": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "tokenizer_file": null,
+  "unk_token": "<unk>",
+  "word_delimiter_token": "|"
+}

fp8/vocab.json

@@ -0,0 +1,42 @@
+{
+  "'": 5,
+  "-": 6,
+  "</s>": 2,
+  "<pad>": 0,
+  "<s>": 1,
+  "<unk>": 3,
+  "|": 4,
+  "а": 7,
+  "б": 8,
+  "в": 9,
+  "г": 10,
+  "д": 11,
+  "е": 12,
+  "ж": 13,
+  "з": 14,
+  "и": 15,
+  "й": 16,
+  "к": 17,
+  "л": 18,
+  "м": 19,
+  "н": 20,
+  "о": 21,
+  "п": 22,
+  "р": 23,
+  "с": 24,
+  "т": 25,
+  "у": 26,
+  "ф": 27,
+  "х": 28,
+  "ц": 29,
+  "ч": 30,
+  "ш": 31,
+  "щ": 32,
+  "ъ": 33,
+  "ы": 34,
+  "ь": 35,
+  "э": 36,
+  "ю": 37,
+  "я": 38,
+  "ё": 39
+}

int4/config.json

@@ -0,0 +1,165 @@
+{
+  "activation_dropout": 0.05,
+  "apply_spec_augment": true,
+  "architectures": [
+    "UniSpeechSatForSequenceClassification"
+  ],
+  "attention_dropout": 0.05,
+  "bos_token_id": 1,
+  "classifier_proj_size": 768,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.05,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.05,
+  "finetuning_task": [
+    "unispeech_sat_classification"
+  ],
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.05,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "anger",
+    "1": "disgust",
+    "2": "enthusiasm",
+    "3": "fear",
+    "4": "happiness",
+    "5": "neutral",
+    "6": "sadness"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "anger": 0,
+    "disgust": 1,
+    "enthusiasm": 2,
+    "fear": 3,
+    "happiness": 4,
+    "neutral": 5,
+    "sadness": 6
+  },
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "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": "unispeech-sat",
+  "num_attention_heads": 16,
+  "num_clusters": 504,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "num_negatives": 100,
+  "pad_token_id": 0,
+  "pooling_mode": "mean",
+  "problem_type": "single_label_classification",
+  "proj_codevector_dim": 768,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "pack-quantized",
+        "input_activations": null,
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 128,
+          "num_bits": 4,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "group",
+          "symmetric": false,
+          "type": "int",
+          "zp_dtype": "torch.int8"
+        }
+      }
+    },
+    "format": "pack-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "unispeech_sat.feature_projection.projection",
+      "projector",
+      "classifier"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.17.1"
+  },
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "transformers_version": "5.10.1",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 40,
+  "xvector_output_dim": 512
+}

int4/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1796bf3fc97b4dab77d52b621a66a6148da582f7f2cf89d0e5226865c63d79b5
+size 218676540

int4/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": true,
+  "sampling_rate": 16000
+}

int4/recipe.yaml

@@ -0,0 +1,9 @@
+default_stage:
+  default_modifiers:
+    QuantizationModifier:
+      targets: [Linear]
+      ignore: ['re:.*feature_extractor.*', 're:.*feature_projection.*', 're:.*pos_conv_embed.*',
+        're:.*gru_rel_pos_linear$', projector, classifier, 're:.*audio_projector$', 're:.*text_projector$',
+        're:.*fusion_module.*', 're:.*text_model\.pooler.*']
+      scheme: W4A16_ASYM
+      bypass_divisibility_checks: false

int4/special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

int4/tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "bos_token": "<s>",
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "Ar4ikov/unispeech-sat-projector-russian-resd",
+  "pad_token": "<pad>",
+  "processor_class": "Wav2Vec2Processor",
+  "replace_word_delimiter_char": " ",
+  "special_tokens_map_file": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "tokenizer_file": null,
+  "unk_token": "<unk>",
+  "word_delimiter_token": "|"
+}

int4/vocab.json

@@ -0,0 +1,42 @@
+{
+  "'": 5,
+  "-": 6,
+  "</s>": 2,
+  "<pad>": 0,
+  "<s>": 1,
+  "<unk>": 3,
+  "|": 4,
+  "а": 7,
+  "б": 8,
+  "в": 9,
+  "г": 10,
+  "д": 11,
+  "е": 12,
+  "ж": 13,
+  "з": 14,
+  "и": 15,
+  "й": 16,
+  "к": 17,
+  "л": 18,
+  "м": 19,
+  "н": 20,
+  "о": 21,
+  "п": 22,
+  "р": 23,
+  "с": 24,
+  "т": 25,
+  "у": 26,
+  "ф": 27,
+  "х": 28,
+  "ц": 29,
+  "ч": 30,
+  "ш": 31,
+  "щ": 32,
+  "ъ": 33,
+  "ы": 34,
+  "ь": 35,
+  "э": 36,
+  "ю": 37,
+  "я": 38,
+  "ё": 39
+}

int8/config.json

@@ -0,0 +1,165 @@
+{
+  "activation_dropout": 0.05,
+  "apply_spec_augment": true,
+  "architectures": [
+    "UniSpeechSatForSequenceClassification"
+  ],
+  "attention_dropout": 0.05,
+  "bos_token_id": 1,
+  "classifier_proj_size": 768,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.05,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.05,
+  "finetuning_task": [
+    "unispeech_sat_classification"
+  ],
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.05,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "anger",
+    "1": "disgust",
+    "2": "enthusiasm",
+    "3": "fear",
+    "4": "happiness",
+    "5": "neutral",
+    "6": "sadness"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "anger": 0,
+    "disgust": 1,
+    "enthusiasm": 2,
+    "fear": 3,
+    "happiness": 4,
+    "neutral": 5,
+    "sadness": 6
+  },
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "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": "unispeech-sat",
+  "num_attention_heads": 16,
+  "num_clusters": 504,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "num_negatives": 100,
+  "pad_token_id": 0,
+  "pooling_mode": "mean",
+  "problem_type": "single_label_classification",
+  "proj_codevector_dim": 768,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "pack-quantized",
+        "input_activations": null,
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 128,
+          "num_bits": 8,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "group",
+          "symmetric": true,
+          "type": "int",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "pack-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "unispeech_sat.feature_projection.projection",
+      "projector",
+      "classifier"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.17.1"
+  },
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "transformers_version": "5.10.1",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 40,
+  "xvector_output_dim": 512
+}

int8/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9456bed8f72193adc478396a1cc7acbf35c3bab05b40083c0510b6cf1c5b8492
+size 368471492

int8/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": true,
+  "sampling_rate": 16000
+}

int8/recipe.yaml

@@ -0,0 +1,9 @@
+default_stage:
+  default_modifiers:
+    QuantizationModifier:
+      targets: [Linear]
+      ignore: ['re:.*feature_extractor.*', 're:.*feature_projection.*', 're:.*pos_conv_embed.*',
+        're:.*gru_rel_pos_linear$', projector, classifier, 're:.*audio_projector$', 're:.*text_projector$',
+        're:.*fusion_module.*', 're:.*text_model\.pooler.*']
+      scheme: W8A16
+      bypass_divisibility_checks: false

int8/special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

int8/tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "bos_token": "<s>",
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "Ar4ikov/unispeech-sat-projector-russian-resd",
+  "pad_token": "<pad>",
+  "processor_class": "Wav2Vec2Processor",
+  "replace_word_delimiter_char": " ",
+  "special_tokens_map_file": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "tokenizer_file": null,
+  "unk_token": "<unk>",
+  "word_delimiter_token": "|"
+}

int8/vocab.json

@@ -0,0 +1,42 @@
+{
+  "'": 5,
+  "-": 6,
+  "</s>": 2,
+  "<pad>": 0,
+  "<s>": 1,
+  "<unk>": 3,
+  "|": 4,
+  "а": 7,
+  "б": 8,
+  "в": 9,
+  "г": 10,
+  "д": 11,
+  "е": 12,
+  "ж": 13,
+  "з": 14,
+  "и": 15,
+  "й": 16,
+  "к": 17,
+  "л": 18,
+  "м": 19,
+  "н": 20,
+  "о": 21,
+  "п": 22,
+  "р": 23,
+  "с": 24,
+  "т": 25,
+  "у": 26,
+  "ф": 27,
+  "х": 28,
+  "ц": 29,
+  "ч": 30,
+  "ш": 31,
+  "щ": 32,
+  "ъ": 33,
+  "ы": 34,
+  "ь": 35,
+  "э": 36,
+  "ю": 37,
+  "я": 38,
+  "ё": 39
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4974bb17eb3f6e3e464448c05cf5ca9ef3715c5961063e68a5a9c92d183e7372
+size 1264964820

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": true,
+  "sampling_rate": 16000
+}

special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "bos_token": "<s>",
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "Ar4ikov/unispeech-sat-projector-russian-resd",
+  "pad_token": "<pad>",
+  "processor_class": "Wav2Vec2Processor",
+  "replace_word_delimiter_char": " ",
+  "special_tokens_map_file": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "tokenizer_file": null,
+  "unk_token": "<unk>",
+  "word_delimiter_token": "|"
+}

vocab.json

@@ -0,0 +1,42 @@
+{
+  "'": 5,
+  "-": 6,
+  "</s>": 2,
+  "<pad>": 0,
+  "<s>": 1,
+  "<unk>": 3,
+  "|": 4,
+  "а": 7,
+  "б": 8,
+  "в": 9,
+  "г": 10,
+  "д": 11,
+  "е": 12,
+  "ж": 13,
+  "з": 14,
+  "и": 15,
+  "й": 16,
+  "к": 17,
+  "л": 18,
+  "м": 19,
+  "н": 20,
+  "о": 21,
+  "п": 22,
+  "р": 23,
+  "с": 24,
+  "т": 25,
+  "у": 26,
+  "ф": 27,
+  "х": 28,
+  "ц": 29,
+  "ч": 30,
+  "ш": 31,
+  "щ": 32,
+  "ъ": 33,
+  "ы": 34,
+  "ь": 35,
+  "э": 36,
+  "ю": 37,
+  "я": 38,
+  "ё": 39
+}