Upload prediction_artifacts/training_args.yaml with huggingface_hub

prediction_artifacts/training_args.yaml

@@ -0,0 +1,19 @@
+cutoff_len: 1024
+dataset_dir: data
+do_predict: true
+eval_dataset: english_to_kannada_test
+finetuning_type: lora
+flash_attn: auto
+max_new_tokens: 512
+max_samples: 100000
+model_name_or_path: /content/gemma_3_1b_it_kn_ext_init
+output_dir: saves/Gemma-3-1B-Instruct/lora/gemma_3_1b_it_kn_ext_init
+per_device_eval_batch_size: 16
+predict_with_generate: true
+preprocessing_num_workers: 16
+quantization_method: bnb
+stage: sft
+temperature: 0.95
+template: gemma
+top_p: 0.7
+trust_remote_code: true