-
Open this to track different decoding strategies of the transducer model. I will update some materials and code snippets here
Suggestions welcome and hope to collaborate
Mddct updated
9 months ago
-
Reporting some funky behavior with the pointer-generator with features:
1. With `--arch pointer_generator_lstm` and features enabled (i.e., `--features_col 3` or something other than the default `0…
-
**Describe the bug 🐞**
The trajectories of the N-ODE solution cut each other in the ODE solution.
**Expected behavior**
As of my understanding, N-ODE trajectories for different starting value…
-
### Before reporting an issue
- [X] I have read and understood the above terms for submitting issues, and I understand that my issue may be closed without action if I do not follow them.
### Area
a…
-
## Environment
- Development Kit: custom board
- Core (if using chip or module): ESP32-D0WDQ6
- IDF version: master
- Development Env: C++ / arduino as idf component
- Power Supply: exter…
-
**Describe the bug**
I am trying to finetune the XXL Conformer-Transducer model from a base .nemo checkpoint that I downloaded from https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_…
-
I conducted several experiments using ctc, rnn-t, and attention loss on zipformer. Attach WER results on librispeech test clean and test other here for reference.
| Exp | Attention loss scale, CT…
-
在训练30epoch后，进行stage5进行解码的时候，没有任何输出结果，也没有报错，直接停掉了，改小bs也没有解决，请问如何解决？
日志如下:
/rnnt$ bash run.sh --stage 5 --stop-stage 5
do model average and final checkpoint is exp/conformer_rnnt/avg_3.pt
Namesp…
-
I am simply re-running the code discussed in https://github.com/SciML/SciMLSensitivity.jl/issues/938 and implemented as a test here https://github.com/SciML/SciMLSensitivity.jl/blob/d1ded754fe3b0d505d…
-
Hello! I have a problem when I train ASR models from scratch on my arabic dataset
when I use pretrained fast-conformer model, It works well and converges
![image](https://github.com/NVIDIA/NeMo/…