Model Card for Model Llama-speechLMM_v1.0_L-ASR
Model information
This is the version of meetween/Llama-speechlmm-1.0-l that was fine-tuned for Automatic Speech Recognition.
Model Architecture
Identical to base model. This model does not include a video adapter.
This model was obtained by fine-tuning the speech adapter. This repository contains the weights of the
speech adapter together with the weights of the main model.
How to use
Identical to main model
Training Data
The model was fine tuned on the same data sets used for training
the main model.
Number of samples (hours):
500 (LibriHeavy) +
300 (LibriTTS) +
68 (AMI) +
68 (ICSI) +
780 (CoVoST2) +
264 (CommonVoice)
= 1980 in total
Evaluation results (%WER)
| ACL6060 (en) | Covost (it) | Covost (es) | Covost (fr) | Covost (de) |
|---|
| Base model | 24.52 | 8.60 | 7.14 | 11.96 | 9.42 |
| SpeechLMM_v1.0_L_FT | 16.52 | 5.65 | 5.17 | 9.44 | 6.57 |
Framework versions
Transformers 4.45.0
Pytorch 2.3.1+cu124.post2
Datasets 3.2.0
Tokenizers 0.20.0
