Mula-4x160-v0.1 / README.md

Update README.md

3ad6cba verified 25 days ago

No virus

12 kB

	---
	language:
	- pt
	license: apache-2.0
	library_name: transformers
	tags:
	- text-generation-inference
	datasets:
	- nicholasKluge/Pt-Corpus-Instruct
	metrics:
	- perplexity
	pipeline_tag: text-generation
	widget:
	- text: A PUCRS é uma universidade
	example_title: Exemplo
	- text: A muitos anos atrás, em uma galáxia muito distante, vivia uma raça de
	example_title: Exemplo
	- text: Em meio a um escândalo, a frente parlamentar pediu ao Senador Silva para
	example_title: Exemplo
	inference:
	parameters:
	repetition_penalty: 1.2
	temperature: 0.2
	top_k: 20
	top_p: 0.2
	max_new_tokens: 150
	co2_eq_emissions:
	emissions: 7.6
	source: CodeCarbon
	training_type: pre-training
	geographical_location: Germany
	hardware_used: NVIDIA A100-SXM4-40GB
	model-index:
	- name: Mula-4x160-v0.1
	results:
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: ENEM Challenge (No Images)
	type: eduagarcia/enem_challenge
	split: train
	args:
	num_few_shot: 3
	metrics:
	- type: acc
	value: 21.34
	name: accuracy
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: BLUEX (No Images)
	type: eduagarcia-temp/BLUEX_without_images
	split: train
	args:
	num_few_shot: 3
	metrics:
	- type: acc
	value: 25.17
	name: accuracy
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: OAB Exams
	type: eduagarcia/oab_exams
	split: train
	args:
	num_few_shot: 3
	metrics:
	- type: acc
	value: 25.06
	name: accuracy
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: Assin2 RTE
	type: assin2
	split: test
	args:
	num_few_shot: 15
	metrics:
	- type: f1_macro
	value: 33.57
	name: f1-macro
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: Assin2 STS
	type: eduagarcia/portuguese_benchmark
	split: test
	args:
	num_few_shot: 15
	metrics:
	- type: pearson
	value: 11.35
	name: pearson
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: FaQuAD NLI
	type: ruanchaves/faquad-nli
	split: test
	args:
	num_few_shot: 15
	metrics:
	- type: f1_macro
	value: 43.97
	name: f1-macro
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: HateBR Binary
	type: ruanchaves/hatebr
	split: test
	args:
	num_few_shot: 25
	metrics:
	- type: f1_macro
	value: 41.5
	name: f1-macro
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: PT Hate Speech Binary
	type: hate_speech_portuguese
	split: test
	args:
	num_few_shot: 25
	metrics:
	- type: f1_macro
	value: 22.99
	name: f1-macro
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: tweetSentBR
	type: eduagarcia/tweetsentbr_fewshot
	split: test
	args:
	num_few_shot: 25
	metrics:
	- type: f1_macro
	value: 11.24
	name: f1-macro
	source:
	url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=MulaBR/Mula-4x160-v0.1
	name: Open Portuguese LLM Leaderboard
	---
	# Mula-4x160-v0.1

	<img src="./logo-no-bg.png" alt="Mula" height="200">

	## Model Summary

	Mula is a series of Sparse Mixture of Experts (SMoE) language models, all trained natively in Brazilian Portuguese, designed to help democratize LLMs for low-resource languages.

	Mula-4x160-v0.1 is one of our first experiments on pre-training a SMoE, using the [Pt-Corpus-Instruct](https://huggingface.co/datasets/nicholasKluge/Pt-Corpus-Instruct) dataset. It has 4 experts per layer and activates 2 for each token.

	Future versions of Mula will be trained on an extensively larger Brazilian Portuguese dataset.

	## Details

	- Architecture: a Sparse Mixture of Experts (Mixtral implementation) pre-trained via causal language modeling
	- Size: 407,820,288 parameters (only 237,950,976 activated parameters during runtime)
	- Context length: 2048 tokens
	- Dataset: [Pt-Corpus Instruct](https://huggingface.co/datasets/nicholasKluge/Pt-Corpus-Instruct) (6.2B tokens)
	- Language: Portuguese
	- Training time: ~ 30 hours
	- Emissions: 7.6 KgCO2eq (Germany)
	- Total energy consumption: 15 kWh

	## Intended Uses

	The primary intended use of Mula-4x160-v0.1 is to research the challenges related to developing language models for low-resource languages. Checkpoints saved during training are intended to provide a controlled setting for performing scientific experiments. You may also further fine-tune and adapt Mula-4x160-v0.1 for deployment, as long as your use is following the Apache 2.0 license. If you decide to use pre-trained Mula-4x160-v0.1 as a basis for your fine-tuned model, please conduct your own risk and bias assessment.

	## Out-of-scope Use

	Mula-4x160-v0.1 is not intended for deployment. It is not a product and should not be used for human-facing interactions.

	Mula-4x160-v0.1 models are Brazilian Portuguese language only and are not suitable for translation or generating text in other languages.

	Mula-4x160-v0.1 has not been fine-tuned for downstream contexts in which language models are commonly deployed.

	## Basic usage

	Using the `pipeline`:

	```python
	from transformers import pipeline

	generator = pipeline("text-generation", model="MulaBR/Mula-4x160-v0.1")

	completions = generator("Astronomia é a ciência", num_return_sequences=2, max_new_tokens=100)

	for comp in completions:
	print(f"🤖 {comp['generated_text']}")
	```

	Using the `AutoTokenizer` and `AutoModelForCausalLM`:

	```python
	from transformers import AutoTokenizer, AutoModelForCausalLM
	import torch

	# Load model and the tokenizer
	tokenizer = AutoTokenizer.from_pretrained("MulaBR/Mula-4x160-v0.1", revision='main')
	model = AutoModelForCausalLM.from_pretrained("MulaBR/Mula-4x160-v0.1", revision='main')

	# Pass the model to your device
	device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

	model.eval()
	model.to(device)

	# Tokenize the inputs and pass them to the device
	inputs = tokenizer("Astronomia é a ciência", return_tensors="pt").to(device)

	# Generate some text
	completions = model.generate(**inputs, num_return_sequences=2, max_new_tokens=100)

	# Print the generated text
	for i, completion in enumerate(completions):
	print(f'🤖 {tokenizer.decode(completion)}')
	```

	## Limitations

	Like almost all other language models trained on large text datasets scraped from the web, Mula-4x160-v0.1 exhibits behavior that does not make them an out-of-the-box solution to many real-world applications, especially those requiring factual, reliable, nontoxic text generation. Our models are all subject to the following:

	- Hallucinations: This model can produce content that can be mistaken for truth but is, in fact, misleading or entirely false, i.e., hallucination.

	- Biases and Toxicity: This model inherits the social and historical stereotypes from the data used to train it. Given these biases, the model can produce toxic content, i.e., harmful, offensive, or detrimental to individuals, groups, or communities.

	- Unreliable Code: The model may produce incorrect code snippets and statements. These code generations should not be treated as suggestions or accurate solutions.

	- Language Limitations: The model is primarily designed to understand standard Brazilian Portuguese. Other languages might challenge its comprehension, leading to potential misinterpretations or errors in response.

	- Repetition and Verbosity: The model may get stuck on repetition loops (especially if the repetition penalty during generations is set to a meager value) or produce verbose responses unrelated to the prompt it was given.

	Hence, even though our models are released with a permissive license, we urge users to perform their risk analysis on these models if intending to use them for real-world applications and also have humans moderating the outputs of these models in applications where they will interact with an audience, guaranteeing users are always aware they are interacting with a language model.

	## Benchmarks

	Evaluations on benchmarks were performed using the [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) (by [EleutherAI](https://www.eleuther.ai/)). [Laiviet](https://github.com/laiviet/lm-evaluation-harness) translated the tasks from the LM-Evaluation-Harness we used.

	\| \| ARC \| HellaSwag \| MMLU \| TruthfulQA \|
	\|----------------------\|-----------\|---------------\|-----------\|----------------\|
	\| Mula-4x160-v0.1 \| 27.09 \| 31.41 \| 28.15 \| 39.81 \|
	\| Mula-8x160-v0.1 \| 26.15 \| 33.06 \| 28.14 \| 41.69 \|

	Evaluations on Brazilian Portuguese benchmarks were performed using a [Portuguese implementation of the EleutherAI LM Evaluation Harness](https://github.com/eduagarcia/lm-evaluation-harness-pt) (created by [Eduardo Garcia](https://github.com/eduagarcia/lm-evaluation-harness-pt)).

	\| \| ASSIN2 RTE \| ASSIN2 STS \| BLUEX \| ENEM \| FAQUAD NLI \| HateBR \| PT Hate Speech \| OAB Exams \| TweetSentBR \|
	\|-----------------------\|----------------\|----------------\|-----------\|----------\|----------------\|------------\|--------------------\|---------------\|-----------------\|
	\| Mula-4x160-v0.1 \| 33.57 \| 11.35 \| 25.17 \| 21.34 \| 43.97 \| 41.50 \| 22.99 \| 25.06 \| 11.24 \|
	\| Mula-8x160-v0.1 \| 33.51 \| 0 \| 20.17 \| 19.94 \| 43.97 \| 33.33 \| 42.69 \| 24.37 \| 24.60 \|

	## Cite as 🤗

	```latex

	@misc{mula2024BR,
	title = {Mula: a Sparse Mixture of Experts Language Model trained in Brazilian Portuguese},
	author = {Corr{\^e}a, Nicholas Kluge and Sen, Aniket and Falk, Sophia and Fatimah, Shiza},
	howpublished = {\url{https://huggingface.co/MulaBR}},
	year={2024}
	}

	```

	## License

	Mula-4x160-v0.1 is licensed under the Apache License, Version 2.0. See the [LICENSE](LICENSE) file for more details.

	## Acknowledgements

	The authors gratefully acknowledge the granted access to the [Marvin cluster](https://www.hpc.uni-bonn.de/en/systems/marvin) hosted by the [University of Bonn](https://www.uni-bonn.de/en) along with the support provided by its High Performance Computing & Analytics Lab.