Downloads · 30 days
6
12% of all-time downloads
Vidhansavaliya123/dlops-ass5-q1-vit-lora-best
dlops-ass5-q1-vit-lora-best is a image classification model from Vidhansavaliya123. Use it when you need a label for an image. It is set up for peft. The card lists the license as mit.
Assignment: DLOps Assignment 5 - Q1 Author: Vidhan Savaliya (M25CSA031, IIT Jodhpur) Base model: timm/vitsmallpatch16224 pre-trained on ImageNet Fine-tuning: LoRA (PEFT) on fused Q/K/V attention projections
Downloads · 30 days
6
12% of all-time downloads
All-time downloads
49
Public
Repo size
452 KB
Likes
0
Public
Click a slice to open those files.
.safetensors452 KB · 99%
From the Hugging Face model README
Assignment: DLOps Assignment 5 - Q1
Author: Vidhan Savaliya (M25CSA031, IIT Jodhpur)
Base model: timm/vit_small_patch16_224 pre-trained on ImageNet
Fine-tuning: LoRA (PEFT) on fused Q/K/V attention projections
| Parameter | Value |
|---|---|
| Rank (r) | 4 |
| Alpha | 4 |
| Dropout | 0.1 |
| Target modules | attn.qkv (all attention blocks) |
| Trainable params | 112,228 / 21,816,392 (0.51%) |
| Split | Accuracy |
|---|---|
| Train (epoch 9) | 93.70% |
| Validation (best) | 90.88% |
| Test | 90.88% |
Baseline (head-only): 78.12% -> +12.76% improvement with LoRA
https://wandb.ai/vidhan-savaliya-indian-institute-of-technology-jodhpur/dlops-ass5-q1/overview
import timm
from peft import PeftModel
base = timm.create_model('vit_small_patch16_224', pretrained=False, num_classes=100)
model = PeftModel.from_pretrained(base, 'Vidhansavaliya123/dlops-ass5-q1-vit-lora-best')
model.eval()