Shahriar/SoACer

# SoACer

This Hugging Face repository contains the best‐performing 
`DownstreamModelSingle` classifier head trained on SoAC embeddings.

## Model Files
 - `pytorch_model.bin`: PyTorch `state_dict()` of the classifier head.
 - `config.json`: JSON with model hyperparameters (embed_size, common_dim, etc.).
 - `README.md`: Usage instructions.

## How to load

```python
import torch
from DownstreamModelSingle import DownstreamModelSingle

# 1) Clone the repo
#    git clone https://huggingface.co/Shahriar/SoACer

# 2) Load config.json
import json
cfg = json.load(open("SoACer/config.json"))

model = DownstreamModelSingle(
    embed_size=cfg["embed_size"],
    class_num=cfg["class_num"],
    common_dim=cfg["common_dim"]
)

# 3) Load weights
state_dict = torch.load("SoACer/pytorch_model.bin", map_location="cpu")
model.load_state_dict(state_dict)
model.eval()

# 4) Use `model` on new precomputed embeddings:
#    given `new_embedding`: a 1D Tensor of size (embed_size,)
#    logits = model(new_embedding.unsqueeze(0))  # shape (1, class_num)
#    probs = torch.softmax(logits, dim=-1)
# ```