Aynkader/translategemma-balochi-adapter

TranslateGemma Balochi Latin LoRA (adapter)

LoRA adapter fine-tuned on Balochi Latin (Syáhag) ↔ English for google/translategemma-4b-it.

• Checkpoint: 8100 (best eval_loss ≈ 0.436)
• Script: Latin Syáhag (á/é/ó, digraphs Ch/Dh/Sh/Th/Zh)
• Owner: Aynkader
• Demo Space: Aynkader/rajank

Load

from peft import PeftModel
from transformers import AutoModelForImageTextToText, AutoProcessor
import torch

base = "google/translategemma-4b-it"
adapter = "Aynkader/translategemma-balochi-adapter"

processor = AutoProcessor.from_pretrained(base, token=True)
model = AutoModelForImageTextToText.from_pretrained(
    base, torch_dtype=torch.bfloat16, device_map="auto", token=True
)
model = PeftModel.from_pretrained(model, adapter)

Accept the base model license and set HF_TOKEN for gated access.

Related

• Training tree: Aynkader/translategemma-balochi-latin