--- language: - pt license: apache-2.0 library_name: transformers tags: - text-generation-inference datasets: - nicholasKluge/instruct-aira-dataset-v3 - cnmoro/GPT4-500k-Augmented-PTBR-Clean - rhaymison/orca-math-portuguese-64k - nicholasKluge/reward-aira-dataset metrics: - perplexity pipeline_tag: text-generation widget: - text: "Cite algumas bandas de rock brasileiras famosas." example_title: Exemplo - text: "Invente uma história sobre um encanador com poderes mágicos." example_title: Exemplo - text: "Qual cidade é a capital do estado do Rio Grande do Sul?" example_title: Exemplo - text: "Diga o nome de uma maravilha culinária característica da cosinha Portuguesa?" 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: 42270 source: CodeCarbon training_type: pre-training geographical_location: Germany hardware_used: NVIDIA A100-SXM4-80GB model-index: - name: Tucano-2b4-Instruct results: - task: type: text-generation name: Text Generation dataset: name: CALAME-PT type: NOVA-vision-language/calame-pt split: all args: num_few_shot: 0 metrics: - type: acc value: 57.66 name: accuracy source: url: https://huggingface.co/datasets/NOVA-vision-language/calame-pt name: Context-Aware LAnguage Modeling Evaluation for Portuguese - task: type: text-generation name: Text Generation dataset: name: LAMBADA-PT type: TucanoBR/lambada-pt split: train args: num_few_shot: 0 metrics: - type: acc value: 39.92 name: accuracy source: url: https://huggingface.co/datasets/TucanoBR/lambada-pt name: LAMBADA-PT - 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: 20.43 name: accuracy source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard 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: 22.81 name: accuracy source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard 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: 24.83 name: accuracy source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard 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: 43.39 name: f1-macro source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard 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: 10 metrics: - type: pearson value: 6.31 name: pearson source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard 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 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: 27.70 name: f1-macro source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard 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: 29.18 name: f1-macro source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard name: Open Portuguese LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: tweetSentBR type: eduagarcia-temp/tweetsentbr split: test args: num_few_shot: 25 metrics: - type: f1_macro value: 43.11 name: f1-macro source: url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard name: Open Portuguese LLM Leaderboard - task: type: text-generation name: Text Generation dataset: name: ARC-Challenge (PT) type: arc_pt args: num_few_shot: 25 metrics: - type: acc_norm value: 32.05 name: normalized accuracy source: url: https://github.com/nlp-uoregon/mlmm-evaluation name: Evaluation Framework for Multilingual Large Language Models - task: type: text-generation name: Text Generation dataset: name: HellaSwag (PT) type: hellaswag_pt args: num_few_shot: 10 metrics: - type: acc_norm value: 48.28 name: normalized accuracy source: url: https://github.com/nlp-uoregon/mlmm-evaluation name: Evaluation Framework for Multilingual Large Language Models - task: type: text-generation name: Text Generation dataset: name: TruthfulQA (PT) type: truthfulqa_pt args: num_few_shot: 0 metrics: - type: mc2 value: 38.44 name: bleurt source: url: https://github.com/nlp-uoregon/mlmm-evaluation name: Evaluation Framework for Multilingual Large Language Models - task: type: text-generation name: Text Generation dataset: name: Alpaca-Eval (PT) type: alpaca_eval_pt args: num_few_shot: 0 metrics: - type: lc_winrate value: 13.0 name: length controlled winrate source: url: https://github.com/tatsu-lab/alpaca_eval name: AlpacaEval base_model: - TucanoBR/Tucano-2b4 --- # Tucano-2b4-Instruct An illustration of a Tucano bird showing vibrant colors like yellow, orange, blue, green, and black. ## Model Summary Tucano-2b4-Instruct is a fine-tuned version of [Tucano-2b4](https://huggingface.co/TucanoBR/Tucano-2b4). **[Tucano](https://huggingface.co/TucanoBR)** is a series of decoder-transformers natively pretrained in Portuguese. All Tucano models were trained on **[GigaVerbo](https://huggingface.co/datasets/TucanoBR/GigaVerbo)**, a concatenation of deduplicated Portuguese text corpora amounting to 200 billion tokens. The fine-tuning process was divided into two stages: - Supervised fine-tuning (SFT) using the [TucanoBR/Tucano-SFT](https://huggingface.co/datasets/TucanoBR/Tucano-SFT), a concatenation of three different instruction tuning datasets ([`cnmoro/GPT4-500k-Augmented-PTBR-Clean`](https://huggingface.co/datasets/cnmoro/GPT4-500k-Augmented-PTBR-Clean), [`rhaymison/orca-math-portuguese-64k`](https://huggingface.co/datasets/rhaymison/orca-math-portuguese-64k), [`nicholasKluge/instruct-aira-dataset-v3`](https://huggingface.co/datasets/nicholasKluge/instruct-aira-dataset-v3)). - Direct Preference Optimization (DPO) using the [nicholasKluge/reward-aira-dataset](https://huggingface.co/datasets/nicholasKluge/reward-aira-dataset). Read our preprint [here](https://arxiv.org/abs/2411.07854). ## Details - **Architecture:** a Transformer-based model pre-trained via causal language modeling - **Size:** 2,444,618,240 parameters - **Context length:** 4096 tokens - **Dataset:** - [cnmoro/GPT4-500k-Augmented-PTBR-Clean](https://huggingface.co/datasets/cnmoro/GPT4-500k-Augmented-PTBR-Clean) - [rhaymison/orca-math-portuguese-64k](https://huggingface.co/datasets/rhaymison/orca-math-portuguese-64k) - [nicholasKluge/instruct-aira-dataset-v3](https://huggingface.co/datasets/nicholasKluge/instruct-aira-dataset-v3) - [nicholasKluge/reward-aira-dataset](https://huggingface.co/datasets/nicholasKluge/reward-aira-dataset) - **Language:** Portuguese - **Training time**: ~ 12 hours - **Emissions:** 42 KgCO2 (Germany) - **Total energy consumption:** 110 kWh This repository has the [source code](https://github.com/Nkluge-correa/Tucano) used to train this model. The main libraries used are: - [PyTorch](https://github.com/pytorch/pytorch) - [Transformers](https://github.com/huggingface/transformers) - [Datasets](https://github.com/huggingface/datasets) - [Tokenizers](https://github.com/huggingface/tokenizers) - [Sentencepiece](https://github.com/google/sentencepiece) - [Accelerate](https://github.com/huggingface/accelerate) - [FlashAttention](https://github.com/Dao-AILab/flash-attention) - [Liger Kernel](https://github.com/linkedin/Liger-Kernel) - [Codecarbon](https://github.com/mlco2/codecarbon) - [TRL](https://github.com/huggingface/trl) ## Intended Uses The primary intended use of the Tucano models is to serve as foundations for research and development involving native Portuguese language modeling. Checkpoints saved during training are designed to provide a controlled setting for performing comparative experiments, specifically regarding the effects of active pretraining on the performance of currently available benchmarks. You may also fine-tune and adapt Tucano models for deployment if your use follows the Apache 2.0 license. If you decide to use the Tucano models as a basis for your fine-tuned model, please conduct your own risk and bias assessment. ## Out-of-scope Use - Tucano models are **not intended for deployment**. They are not an out-of-the-box product and should not be used for human-facing interactions. - Tucano models are for **the Portuguese language only** and are unsuitable for text generation tasks in other languages. - Tucano models have **not been fine-tuned** for downstream tasks. ## Basic usage Using the `pipeline`: ```python from transformers import pipeline generator = pipeline("text-generation", model="TucanoBR/Tucano-2b4-Instruct") completions = generator("Qual cidade é a capital do estado do Rio Grande do Sul?", 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 GenerationConfig, TextGenerationPipeline, AutoTokenizer, AutoModelForCausalLM import torch # Specify the model and tokenizer model_id = "TucanoBR/Tucano-2b4-Instruct" tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained(model_id) # Specify the generation parameters as you like generation_config = GenerationConfig( **{ "do_sample": True, "max_new_tokens": 2048, "renormalize_logits": True, "repetition_penalty": 1.2, "temperature": 0.3, "top_k": 30, "top_p": 0.3, "use_cache": True, } ) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") generator = TextGenerationPipeline(model=model, task="text-generation", tokenizer=tokenizer, device=device) # Generate text prompt = "Qual cidade é a capital do estado do Rio Grande do Sul?" completion = generator(prompt, generation_config=generation_config) print(completion[0]['generated_text']) ``` ## Limitations Like almost all other language models trained on large text datasets scraped from the web, the Tucano models show behavior that does not make them an out-of-the-box solution to many real-world applications, especially those requiring factual, reliable, and nontoxic text generation. Tucano models are all subject to the following: - **Hallucinations:** Tucano models can produce content that can be mistaken as true facts, but are misleading or entirely false, i.e., hallucination. - **Biases and Toxicity:** Tucano models inherit the social and historical stereotypes from the data used to train them. Given these biases, the model can produce toxic content, i.e., harmful, offensive, or detrimental to individuals, groups, or communities. - **Unreliable Code:** Tucano models may produce incorrect code snippets and statements. These code generations should not be treated as suggestions or accurate solutions. - **Language Limitations:** Tucano models are primarily designed to interact with Portuguese. Other languages might challenge its comprehension, leading to potential misinterpretations or errors in response. - **Repetition and Verbosity:** Tucano models 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 them if they intend to use them for real-world applications. ## Evaluations To evaluate the `Instruct` versions of our models, we used [AlpacaEval](https://github.com/tatsu-lab/alpaca_eval) 2.0 with length-controlled win rates, a fast and relatively cheap evaluation method that is highly correlated with human preferences and evaluations of pairwise comparisons. To learn more about our evaluation read [our documentation](https://github.com/Nkluge-correa/Tucano/blob/main/evaluations/README.md). | | Avg. Length | Wins | Base Wins | Total Matches | Length-Controlled Win Rate (%) | LC Std. Error | |-------------------------|-------------|------|-----------|---------------|--------------------------------|---------------| | Llama-3.2-3B-Instruct | 1609 | 257 | 548 | 805 | 21.06 | 0.075 | | **Tucano-2b4-Instruct** | 1843 | 151 | 654 | 805 | 13.00 | 0.071 | | **Tucano-1b1-Instruct** | 1667 | 124 | 681 | 805 | 8.80 | 0.083 | | Llama-3.2-1B-Instruct | 1429 | 99 | 706 | 805 | 7.15 | 0.057 | | TeenyTinyLlama-460m-Chat| 1333 | 28 | 777 | 805 | 2.84 | 0.059 | | Sabiá-7b | 5011 | 1 | 804 | 805 | 0.076 | 0.0043 | | Gervásio-7b | 5740 | 1 | 804 | 805 | 0.026 | 0.0016 | ## Cite as 🤗 ```latex @misc{correa2024tucanoadvancingneuraltext, title={{Tucano: Advancing Neural Text Generation for Portuguese}}, author={Corr{\^e}a, Nicholas Kluge and Sen, Aniket and Falk, Sophia and Fatimah, Shiza}, year={2024}, eprint={2411.07854}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2411.07854}, } ``` ## Aknowlegments We gratefully acknowledge the granted access to the [Marvin cluster](https://www.hpc.uni-bonn.de/en/systems/marvin) hosted by [University of Bonn](https://www.uni-bonn.de/en) along with the support provided by its High Performance Computing \& Analytics Lab. ## License Tucano is licensed under the Apache License, Version 2.0. For more details, see the [LICENSE](LICENSE) file.