Update README.md

aca7a65 12 months ago

6.73 kB

	---
	license: apache-2.0
	tags:
	- generated_from_trainer
	- alpaca
	- self-instruct
	- instruction generation
	- instructiongen
	datasets:
	- pszemraj/fleece2instructions
	metrics:
	- rouge
	model-index:
	- name: bart-base-instructiongen
	results:
	- task:
	name: Sequence-to-sequence Language Modeling
	type: text2text-generation
	dataset:
	name: pszemraj/fleece2instructions
	type: pszemraj/fleece2instructions
	split: validation
	metrics:
	- name: Rouge1
	type: rouge
	value: 61.7209
	widget:
	- text: >-
	You'll need to start by choosing the right venue. Consider the type of
	atmosphere and the size of the area that will be suitable for the number of
	guests you plan to invite. Choose the right decorations based on your
	brother's interests, such as balloons in his favorite colors, banners, and
	streamers. Next, decide on the food and drinks, making sure they are tasty
	and appropriate for the occasion. Then decide on the other games, music, and
	entertainment that will make the party memorable. Finally, involve your
	brother's friends and family to help create the perfect surprise.
	example_title: birthday party
	- text: 1) cookies and cream 2) chocolate chip 3) mint chip 4) oreo
	example_title: ice cream
	- text: >-
	Start by selecting a scale model of a building that fits the theme. Use a
	hobby knife and glue to cut and assemble the model into a ruined or
	abandoned version of itself, adding details like broken windows and
	graffiti. Create a base for the diorama using foam, plaster, or other
	materials, and paint it to resemble a ruined street or sidewalk. Add
	miniature vehicles, debris, and figures to complete the scene, and use
	weathering techniques like dry brushing and rust washes to add realism.
	Display the diorama in a shadow box or other protective case to showcase
	your work.
	example_title: Miniature diorama creation
	- text: >-
	Start by selecting clothing that is futuristic and edgy, such as leather
	jackets, neon-colored accessories, and tech-inspired patterns. Add
	accessories like goggles, cybernetic implants, and LED lights to enhance the
	cyberpunk vibe. Use makeup and body paint to create a futuristic look, such
	as metallic skin or neon makeup. Consider adding functional elements to your
	costume, such as a built-in backpack or hidden pockets for your tech
	gadgets. Finally, practice your confident walk and embrace your inner
	cyberpunk for a memorable and immersive costume experience.
	example_title: Cyberpunk costume design
	- text: >-
	Start by creating a base terrain with mountains, valleys, and other natural
	features. Use fractal noise and displacement mapping to add texture and
	detail to the terrain, and experiment with different materials like rock,
	grass, and water. Add surreal elements like floating islands, giant
	mushrooms, or impossible geometry to create a dreamlike atmosphere. Use
	lighting and color grading to enhance the mood and tone of the scene, and
	render the final image at a high resolution for maximum impact. Share your
	surreal landscape with the world and inspire others to explore the
	possibilities of 3D art.
	example_title: Surreal 3D landscape creation
	- text: >-
	Start by setting a realistic goal and creating a training plan. Build up
	your mileage gradually over time, and incorporate cross-training and
	strength exercises to prevent injury and improve endurance. Be sure to stay
	hydrated and properly fuel your body with nutritious foods. Listen to your
	body and adjust your training as needed to avoid overexertion or burnout.
	Finally, taper your training in the weeks leading up to the race to give
	your body time to rest and recover before the big day.
	example_title: Marathon training
	inference:
	parameters:
	max_length: 96
	num_beams: 4
	---


	# bart-base-instructiongen

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

	- Check out a [basic demo on Spaces](https://huggingface.co/spaces/pszemraj/generate-instructions)
	- An example of how to use instructiongen models in a CLI script can be found [here](https://gist.github.com/pszemraj/8b0213e700763106074d3ac15d041c14)
	- You can find other models fine-tuned for instruction generation by [searching for the instructiongen tag](https://huggingface.co/models?other=instructiongen).


	## About

	Hypothesis: Apply text-to-text models to unlabeled domain-specific text to generate appropriate LLM instructions. Consequently, this may enable domain adaptation of instruction-tuned LLMs, making them more versatile for specific domains.

	This model is a fine-tuned version of the [facebook/bart-base](https://huggingface.co/facebook/bart-base) model, fine-tuned using the `pszemraj/fleece2instructions` dataset.

	It achieves the following results on the evaluation set:
	- Loss: 1.0034
	- Rouge1: 61.7209
	- Rouge2: 45.0116
	- Rougel: 59.8188
	- Rougelsum: 59.8931
	- Gen Len: 14.3179

	## Intended uses & limitations

	This is just a base model/example. There is likely to be even better performance with larger models (click [here to see other checkpoints](https://huggingface.co/models?other=instructiongen))

	Additionally, this 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 procedure

	### Training hyperparameters

	The following hyperparameters were used during training:
	- learning_rate: 8e-05
	- train_batch_size: 8
	- eval_batch_size: 1
	- seed: 42
	- distributed_type: multi-GPU
	- gradient_accumulation_steps: 8
	- 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 \|
	\|:-------------:\|:-----:\|:----:\|:---------------:\|:-------:\|:-------:\|:-------:\|:---------:\|:-------:\|
	\| 1.2723 \| 1.0 \| 362 \| 1.0325 \| 61.6206 \| 45.1199 \| 59.6467 \| 59.7534 \| 14.0443 \|
	\| 1.0157 \| 2.0 \| 724 \| 1.0034 \| 62.4433 \| 46.0114 \| 60.5355 \| 60.6392 \| 14.1807 \|