Update README.md
Browse files
README.md
CHANGED
|
@@ -27,4 +27,10 @@ Orhan, P., Boubenec, Y., & King, J.-R. (2024). Algebraic structures emerge from
|
|
| 27 |
|
| 28 |
Models are pretrained using HuggingFace's trainer.
|
| 29 |
These models pretraining are often shorter (100,000 steps compared to 400 000) than original pretraining because of resource scarcity.
|
| 30 |
-
In my experience, most emergences I studied had happened before 100 000 steps.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
|
| 28 |
Models are pretrained using HuggingFace's trainer.
|
| 29 |
These models pretraining are often shorter (100,000 steps compared to 400 000) than original pretraining because of resource scarcity.
|
| 30 |
+
In my experience, most emergences I studied had happened before 100 000 steps.
|
| 31 |
+
|
| 32 |
+
Known version compatibility issues for Wav2vec2:
|
| 33 |
+
Some Wav2vec2 models are trained with a torch <=2.0.1, while other are trained with torch>2.1.1
|
| 34 |
+
This can create critical error when loading the model, at the step of loading the Wav2Vec2PositionalConvEmbedding,
|
| 35 |
+
which uses different module for the weight_norm depending on the pretraining version. This is quite unfortunate.
|
| 36 |
+
Consequently I recommend checking for the error message of HuggingFace with from_pretrained(), and changing your torch version if the weights can't be properly loaded.
|