flan-t5-base-instructiongen

Instead of generating questions from text, generate instructions for LLMs!

This model is a fine-tuned version of google/flan-t5-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.0642
• Rouge1: 58.9516
• Rouge2: 41.8006
• Rougel: 56.8249
• Rougelsum: 56.9171
• Gen Len: 13.1493

Intended uses & limitations

Of the three models fine-tuned so far, flan-t5-base is in an awkward position where it has the largest model file size, but not the best performance. I'd recommend looking at the two linked below.

This is just a base FLAN model, and is mostly uploaded for comparison with the FLAN-small and bart-base variants.

Additionally, it was trained on a dataset of only instructions+outputs, with the inputs filtered out. This means that text of 1) cookies and cream 2) chocolate chip 3) mint chip 4) oreo will not get you "Rank the following ice cream flavors: oreo, mint chip, chocolate chip, cookies and cream"

Training and evaluation data

See the linked dataset pszemraj/fleece2instructions - it is a filtered/formatted version of tatsu-lab/alpaca to generate instructions for arbitrary text.

• Some of the API examples are intentionally weird to demonstrate the generalizability of the model.

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 8e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.02
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
1.1939	1.0	362	1.0822	58.1758	40.9388	56.1219	56.2464	13.2592
1.1667	2.0	724	1.0642	58.9516	41.8006	56.8249	56.9171	13.1493