Oleguer

Isn't Conv1D with kernel=1 the same as Linear layer but slower?

1

# ❓ Questions & Help I was wandering whether it makes sense from an efficiency point of view to change the Conv1D layer with kernel 1 from the conformer encoder...

Loss becomes NAN after a while

3

## Environment info - Platform: Ubuntu20.04 - Python version: 3.8.10 - PyTorch version (GPU?): 1.9.0+cu111 (pytorch lightning 1.5.8) - Using GPU in script?: 4x A100 ## Information Model I am...

Why are the error rate accumulated?

5

# ❓ Questions & Help  ## Details Each call of the error rate accumulates the distance and length. Why is that?Is it to have a running average kind of...

Best decoder?

1

# ❓ Questions & Help Hi guys! First of all thank you so much for such an amazing repo :) I'd like to know if you have some insights on...

Could there be a memory leak in the conformer_lstm model?

8

# ❓ Questions & Help I am training using the random sampler and using ddp_sharded strategy, however after some training steps I get a CUDA out-of-memory error. ## Details I...

Is it possible to get timestamps for BeamSearchLSTM-based inference?

3

# ❓ Questions & Help I see that the output can be aligned (provide per-token timestamps) if we use the CTCBeamDecoder, I wonder if I can get timestamps also if...

Same batch in distributed backends

2

Hey guys! I might be mistaken but I think the way the samplers are implemented, if using a distributed backend (such as ddp, ddp-sharded), samples the same examples for all...

Shorter decoding segments when using server/x86

2

**Problem:** I was exploring word-level confidences using the transformer decoders and I noticed the following pattern in the "forward-in-time" decoder:The last half part of the logits sequences the ctc-guessed probabilities...

TODOS

2

Task list to remember remaining features: - [x] Add Python Server - [x] Add C++ Client - [x] Add Image transferring capabilities - [x] Add Basic Readme - [ ]...