Embedding Gemma Transcribe — Model Files
Core ML model
Google DeepMind's multimodal embedding model, which is licensed under the Apache License 2.0.
The model files were changed by The Smart Language before release: EmbeddingGemma 2's audio encoder was fine-tuned, new layers were trained to read speech into words and to punctuate, and the result was quantized and converted to Core ML.
Contents
| File | Size | Description |
|---|---|---|
word_unit_f16.bin |
352 MB | Word decoder units (fp16) |
eg2_audio_l11.mlmodelc/ |
233 MB | Fine-tuned EmbeddingGemma-2 audio encoder (layer 11), Core ML |
reader_P6.mlmodelc/ |
67 MB | Speech-to-words reader (P6) |
scorer_blk_d128_k16_b4_t500.mlmodelc/ |
59 MB | Scoring model (beam search scorer) |
punct_headonly.mlmodelc/ |
6.8 MB | Punctuation model |
reader_words.json |
2.6 MB | Word vocabulary |
reader_P6.json, punct_case.json |
small | Decoding / punctuation metadata |
mel_filters_257x128.f32, hann_320.f32 |
small | Audio feature extraction tables |
manifest.json, installed.json |
small | Download manifest and integrity checksums (SHA-256) |
NOTICE |
small | Attribution and license notice |
Total size: ~715 MB
Source
Original distribution: haku-models.tslfiles.org/eg2demo/
Restore locally
hf download selcukkubur/embedding-gemma-transcribe-models --local-dir ./models
License & attribution
- EmbeddingGemma 2 is licensed under the Apache License, Version 2.0 (https://www.apache.org/licenses/LICENSE-2.0).
- EmbeddingGemma 2's model card asks that deployments adhere to the Gemma Prohibited Use Policy.
- Gemma and EmbeddingGemma are trademarks of Google. This app is not affiliated with, sponsored or endorsed by Google.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support