mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-26 18:24:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

decrease learning-rate to solve the error: RuntimeError: grad_scale is too small, exiting: 5.820766091346741e-11

Browse Source
This commit is contained in:
Triplecq 2024-01-14 12:12:15 -05:00
parent ced8a53cdc
commit 42c152f5cb
1 changed files with 1 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
egs/reazonspeech/ASR/zipformer/train.py
View File

@ -30,7 +30,6 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"
--start-epoch 1 \
--use-fp16 1 \
--exp-dir zipformer/exp \
--full-libri 1 \
--max-duration 1000
# For streaming model training:
@ -41,7 +40,6 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"
--use-fp16 1 \
--exp-dir zipformer/exp \
--causal 1 \
--full-libri 1 \
--max-duration 1000
It supports training with:
@ -330,7 +328,7 @@ def get_parser():
)
parser.add_argument(
"--base-lr", type=float, default=0.045, help="The base learning rate."
"--base-lr", type=float, default=0.035, help="The base learning rate."
)
parser.add_argument(