huggingface / huggingface/diffusers
Confusion about `FrozenDict` in `configuration_utils.py`
- Langage dominant
- Python
- Étoiles
- 34.5k
- Forks
- 7.3k
- Merge moyen
- 3 j 3 h
- PR mergées (30 j)
- 91
Description
I am confused about the design of `FrozenDict` in `configuration_utils.py` and the usage of it.
### 1. Is `FrozenDict` really frozen?
From the code, `FrozenDict` sets `self.__frozen = True` during initialization. It then checks `if hasattr(self, "__frozen") and self.__frozen` in methods like `__setattr__` or `__setitem__` to raise an exception if it's supposed to be frozen. However, in Python, using double underscores (`__`) triggers name mangling, which means `hasattr(self, "__frozen")` couldn't find `__frozen` as it is `_FrozenDict__frozen` actually. This check would always return False, rendering the intended freeze ineffective – `__setattr__` and `__setitem__` can still be used, making the `FrozenDict` not truly frozen.
Is this what we expect?
### 2. If we were to modify `FrozenDict` to be truly immutable, how would we use it as `model.config`?
Typically, we register parameters needed for model initialization as a `FrozenDict` via `register_to_config`. These are often accessed during the forward method with checks like `if self.config.xxx == xxx` to determine execution paths. However, in some models, certain properties of `self.config` might need to be altered after initialization, as seen in [IP-Adapter](https://github.com/huggingface/diffusers/blob/main/src/diffusers/loaders/unet.py#L851)
```py
self.config.encoder_hid_dim_type = "ip_image_proj"
```
Does this contradict the design philosophy of `FrozenDict`?
### 3. If `models.config` could be mutable, how should one go about changing it?
In the example above, should we use `__setattr__` to modify `self.config.encoder_hid_dim_type` or `__setitem__` to add an additional key-value pair to `FrozenDict` which appears more in line with typical dictionary usage, like
```diff
- self.config.encoder_hid_dim_type = "ip_image_proj"
+ self.config['encoder_hid_dim_type'] = "ip_image_proj"
```
I was just wondering if you might have a moment to clarify these points for me?
Guide de contribution
Ouvrir le guide de contribution
Piste de recherche
Start by reading configuration_utils.py, then compare FrozenDict's behavior with the self.config.encoder_hid_dim_type assignment in loaders/unet.py. Clarify whether the intended outcome is immutable configuration, supported mutation, or documentation of current behavior; the issue is done only once that scope and expected behavior are decided.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- python
- Domaine
- machine-learning
- Type d'issue
- Bug
- Difficulté
- 5/5
- Temps estimé
- Plus d'une semaine
- Activité
- À l'abandon
- Clarté
- À clarifier
- Accessibilité débutants
- 20/100