Finetuned CLIP Projection Head (Persian)
This repository contains the file finetune_clip_fa.pt,
a projection head (linear layer) fine-tuned on top of a Persian CLIP model.
Contents
finetune_clip_fa.pt→ PyTorch checkpoint of the projection head.
How to Use
import torch
from huggingface_hub import hf_hub_download
# Repository and filename
REPO_ID = "ahm1378/finetune-clip-fa-head"
FILENAME = "finetune_clip_fa.pt"
# Download checkpoint
ckpt_path = hf_hub_download(
repo_id=REPO_ID,
filename=FILENAME,
repo_type="model"
)
# Load into a linear projection head
state = torch.load(ckpt_path, map_location="cpu")
proj_txt = torch.nn.Linear(512, 512, bias=False)
if "proj_txt" in state:
proj_txt.load_state_dict(state["proj_txt"])
elif "state_dict" in state:
proj_txt.load_state_dict(state["state_dict"])
else:
proj_txt.load_state_dict(state)
proj_txt.eval()
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support