Comments (11)
Hi,
Can you elaborate on which argument you are referring to, is that
ssast/src/finetune/esc50/run_esc_patch.sh
Line 41 in a1a3eec
Thanks!
-Yuan
from ssast.
yes
from ssast.
Can you explain why the value would be 1024?
It seems to me that it changes
Lines 97 to 101 in a1a3eec
and
Lines 132 to 138 in a1a3eec
for both dataloading and model instantiation.
from ssast.
because the process of loading the model file ssast-base-patch-400.pth changes the target_length, the code is shown as below
try:
p_fshape, p_tshape = sd['module.v.patch_embed.proj.weight'].shape[2],sd['module.v.patch_embed.proj.weight'].shape[3]
p_input_fdim, p_input_tdim = sd['module.p_input_fdim'].item(), sd['module.p_input_tdim'].item()
from ssast.
我猜测这也是为什么finetune完之后,做纯推理时load model file会报错的原因。不知道我理解的是否对
from ssast.
can you paste the error code here?
from ssast.
can you paste the error code here?
you can print the p_input_tdim before 156 line of ast_model,you will find the error
from ssast.
I don't have enough time to run it again. The code is a cleaned up version from the development version. It went through a brief test and I guess I did take care of this. So if you already have a error message, that would be very helpful. It might due to something else.
from ssast.
我相信很多人都有同样的问题。因为finetune之后保存的模型,根本没法load进来做纯推理,我也不知道该如何测试训练好的模型的真实性能
from ssast.
Oh I see, yes, that is a known problem. It should be fine if you finetune a pretrained model that has different target_length
, but if you want to take the finetuned model for deployment, you will get an error.
For checking the performance, once you finetune a pretrained model, the script will print out the accuracy (or mAP) and also save the result on disk.
For deploy the model for inference, you will need to fix the bug.
from ssast.
Can you check this: #4
from ssast.
Related Issues (20)
- fine-tuning sample rate HOT 1
- Target length HOT 2
- A question about ESC's AC HOT 2
- about nce loss cal HOT 3
- How to convert fbank features back to audio ? HOT 3
- I want to use frame-level ssast just for frame-level audio token extraction HOT 3
- Stereo audio HOT 1
- Model fails to converge on transfer to audio backtesting problem HOT 5
- speech commands v1 HOT 3
- Extract audio representations for future use HOT 13
- sed HOT 1
- python and cuda version HOT 3
- Host on Huggingface HOT 2
- Which python version are you using? HOT 1
- For discriminative loss, is the true NCE batch size the number of masked patches? HOT 2
- Librispeech960 Pretrained Model HOT 3
- Half precision spectograms - Mixed precision training HOT 1
- The model loaded is not from a torch.nn.Dataparallel object HOT 19
- Embeddings without fine tuning HOT 5
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from ssast.