chbsaikiran/microsoft_phi2_finetuned_on_OpenAssistant_oasst1

Fine-tuned Phi-2 Assistant

This is a fine-tuned version of the Microsoft Phi-2 model, adapted for conversation using QLoRA fine-tuning.

Model Details

• —Base Model: Microsoft Phi-2
• —Fine-tuning Method: QLoRA
• —Training Data: [Your dataset details]
• —Purpose: [Your model's purpose]

Usage

The model can be used through the Gradio interface. Simply type your message and the model will respond.

Examples

[Add some example interactions]

Limitations

[Add any known limitations or biases]

<pre> Step Training Loss 10 1.835300 20 1.824400 30 1.869800 40 1.786500 50 1.783500 60 1.726700 70 1.680100 80 1.655000 90 1.744500 100 1.858900 110 1.692100 120 1.665600 130 1.675300 140 1.734400 150 1.800100 160 1.905800 170 1.717500 180 1.892500 190 1.579900 200 1.753700 210 1.703600 220 1.855600 230 1.689700 240 1.760900 250 1.740300 260 1.743000 270 1.835700 280 1.798900 290 1.754400 300 1.810000 310 1.815800 320 1.786900 330 1.738300 340 1.794300 350 1.717100 360 1.789500 370 1.814700 380 1.682100 390 1.938800 400 1.833400 410 1.828800 420 1.763600 430 1.710400 440 1.859600 450 1.943100 460 1.615100 470 1.904200 480 1.734400 490 1.854800 500 1.894900 TrainOutput(globalstep=500, trainingloss=1.7778642253875732, metrics={'trainruntime': 6199.0179, 'trainsamplespersecond': 1.291, 'trainstepspersecond': 0.081, 'totalflos': 6.53499826176e+16, 'train_loss': 1.7778642253875732}) </pre>