Add Edda q8_0, Silero VAD and punct-da

README.md

@@ -0,0 +1,27 @@
+---
+license: other
+license_name: mixed
+language:
+- da
+tags:
+- whisper.cpp
+- ggml
+- punctuation
+---
+
+# Eckermann models
+
+The model files that the [Eckermann](https://github.com/bhark/eckermann) dictation app downloads on first run. The app pins each file by commit and SHA-256.
+
+| File | What it is | Source | License |
+|---|---|---|---|
+| `ggml-edda-v0.2-q8_0.bin` | Edda v0.2 Danish speech recognition, converted to GGML and quantized to q8_0 | [danish-foundation-models/edda-v0.2](https://huggingface.co/danish-foundation-models/edda-v0.2) at `698f8d0`, by the Alexandra Institute in the CoRal project. Based on [openai/whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) (MIT) | Apache 2.0 |
+| `ggml-silero-v6.2.0.bin` | Silero VAD v6.2.0, unchanged | [ggml-org/whisper-vad](https://huggingface.co/ggml-org/whisper-vad), from [snakers4/silero-vad](https://github.com/snakers4/silero-vad) | MIT |
+| `punct-da/` | Danish punctuation and casing restoration, weights as f16 safetensors | [Alvenir/bert-punct-restoration-da](https://huggingface.co/Alvenir/bert-punct-restoration-da) at `922d302`. Based on [Maltehb/danish-bert-botxo](https://huggingface.co/Maltehb/danish-bert-botxo) by BotXO (CC BY 4.0) | Apache 2.0 |
+
+## How the files were made
+
+- Edda: `models/convert-h5-to-ggml.py` from [whisper.cpp](https://github.com/ggml-org/whisper.cpp) at `231638a`, then `whisper-quantize` with `q8_0`.
+- punct-da: [`bench/punct_model.py`](https://github.com/bhark/eckermann/blob/main/bench/punct_model.py) in the Eckermann repo.
+
+The licenses of the original models apply to these copies.

ggml-edda-v0.2-q8_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:976dd84ac3e4e209d8c77de921d6fd0e42b624470578cd897bef653a97e6b4b7
+size 874188075

ggml-silero-v6.2.0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aa269b785eeb53a82983a20501ddf7c1d9c48e33ab63a41391ac6c9f7fb6987
+size 885098

punct-da/config.json

@@ -0,0 +1,66 @@
+{
+    "_name_or_path": "Maltehb/danish-bert-botxo",
+    "architectures": [
+        "BertForTokenClassification"
+    ],
+    "attention_probs_dropout_prob": 0.1,
+    "classifier_dropout": null,
+    "directionality": "bidi",
+    "gradient_checkpointing": false,
+    "hidden_act": "gelu",
+    "hidden_dropout_prob": 0.1,
+    "hidden_size": 768,
+    "id2label": {
+        "0": "OU",
+        "1": "OO",
+        "2": ".O",
+        "3": "!O",
+        "4": ",O",
+        "5": ".U",
+        "6": "!U",
+        "7": ",U",
+        "8": ":O",
+        "9": ";O",
+        "10": ":U",
+        "11": "'O",
+        "12": "-O",
+        "13": "?O",
+        "14": "?U"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "label2id": {
+        "OU": 0,
+        "OO": 1,
+        ".O": 2,
+        "!O": 3,
+        ",O": 4,
+        ".U": 5,
+        "!U": 6,
+        ",U": 7,
+        ":O": 8,
+        ";O": 9,
+        ":U": 10,
+        "'O": 11,
+        "-O": 12,
+        "?O": 13,
+        "?U": 14
+    },
+    "layer_norm_eps": 1e-12,
+    "max_position_embeddings": 512,
+    "model_type": "bert",
+    "num_attention_heads": 12,
+    "num_hidden_layers": 12,
+    "pad_token_id": 0,
+    "pooler_fc_size": 768,
+    "pooler_num_attention_heads": 12,
+    "pooler_num_fc_layers": 3,
+    "pooler_size_per_head": 128,
+    "pooler_type": "first_token_transform",
+    "position_embedding_type": "absolute",
+    "torch_dtype": "float32",
+    "transformers_version": "4.17.0",
+    "type_vocab_size": 2,
+    "use_cache": true,
+    "vocab_size": 32000
+}

punct-da/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a7af679dcf1e8b53c6f1a917628f1c5237f89ece496bec12799acca7e128dd1
+size 220099774