kailasa-ngpt/IndicF5-Tamil-Finetuned
0
Round 4: 25 epochs, 300 broader Tamil+Sanskrit samples, correct 2545-token vocab
Restore original IndicF5 base vocabulary (2545 tokens)
Round 3: 50 epochs, correct 2545-token vocab, 150 pure Tamil samples, lr=3e-5
Update vocab from expanded dataset
Upload 50-epoch weights (150 pure Tamil samples, lr=3e-5)
Upload vocab.txt with huggingface_hub
Upload first successful Tamil fine-tuned weights (15 epochs, vocabulary restored)
initial commit