| """Configuration for Chest2Vec — a Qwen3-Embedding model (LoRA merged in) for chest |
| radiology report embeddings. |
| |
| Self-contained: the merged encoder weights ship in the repo, so loading needs neither the |
| `chest2vec` package nor a download of the base Qwen3-Embedding weights. Produces a single |
| L2-normalized report embedding (last-token / EOS pooling), Qwen3-Embedding convention. |
| """ |
| from typing import Optional |
| from transformers import PretrainedConfig |
|
|
|
|
| class Chest2VecConfig(PretrainedConfig): |
| model_type = "chest2vec" |
|
|
| def __init__( |
| self, |
| encoder_config: Optional[dict] = None, |
| base_model: str = "Qwen/Qwen3-Embedding-0.6B", |
| hidden_size: int = 1024, |
| default_max_len: int = 512, |
| pooling: str = "last_token", |
| matryoshka_dims: Optional[list] = None, |
| **kwargs, |
| ): |
| super().__init__(**kwargs) |
| self.encoder_config = encoder_config or {} |
| self.base_model = base_model |
| self.hidden_size = hidden_size |
| self.default_max_len = default_max_len |
| self.pooling = pooling |
| self.matryoshka_dims = matryoshka_dims or [] |
|
|
