ptrdvn
/

kakugo-3B-grn

@@ -1,199 +1,244 @@
 ---
-library_name: transformers
-tags: []
----
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## Model Details
-### Model Description
-<!-- Provide a longer summary of what this model is. -->
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-### Model Sources [optional]
-<!-- Provide the basic links for the model. -->
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-## Uses
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-### Direct Use
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-[More Information Needed]
-### Downstream Use [optional]
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-[More Information Needed]
-### Out-of-Scope Use
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-[More Information Needed]
-## Bias, Risks, and Limitations
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-[More Information Needed]
-### Recommendations
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-## How to Get Started with the Model
-Use the code below to get started with the model.
-[More Information Needed]
-## Training Details
-### Training Data
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-[More Information Needed]
-### Training Procedure
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-#### Preprocessing [optional]
-[More Information Needed]
-#### Training Hyperparameters
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-#### Speeds, Sizes, Times [optional]
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-[More Information Needed]
-## Evaluation
-<!-- This section describes the evaluation protocols and provides the results. -->
-### Testing Data, Factors & Metrics
-#### Testing Data
-<!-- This should link to a Dataset Card if possible. -->
-[More Information Needed]
-#### Factors
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-[More Information Needed]
-#### Metrics
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-[More Information Needed]
-### Results
-[More Information Needed]
-#### Summary
-## Model Examination [optional]
-<!-- Relevant interpretability work for the model goes here -->
-[More Information Needed]
-## Environmental Impact
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-## Technical Specifications [optional]
-### Model Architecture and Objective
-[More Information Needed]
-### Compute Infrastructure
-[More Information Needed]
-#### Hardware
-[More Information Needed]
-#### Software
-[More Information Needed]
-## Citation [optional]
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-**BibTeX:**
-[More Information Needed]
-**APA:**
-[More Information Needed]
-## Glossary [optional]
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-[More Information Needed]
-## More Information [optional]
-[More Information Needed]
-## Model Card Authors [optional]
-[More Information Needed]
-## Model Card Contact
-[More Information Needed]

 ---
+license: apache-2.0
+datasets:
+- ptrdvn/kakugo-grn
+language:
+- grn
+base_model:
+- ibm-granite/granite-4.0-micro
+pipeline_tag: text-generation
+tags:
+- low-resource-language
+- data-distillation
+- conversation
+- grn
+- Guarani
+---
+# Kakugo 3B Guarani
+[[Paper]](https://arxiv.org/abs/2601.14051) [[Code]](https://github.com/Peter-Devine/kakugo) [[Dataset]](https://huggingface.co/datasets/ptrdvn/kakugo-grn)
+<div align="center">
+    <div style="font-size: 80px;font-family: Arial, Helvetica, sans-serif;font-variant: small-caps;color: #000000;font-weight: 700; margin-top:-40px; margin-bottom:-60px; margin-left: -20px" align="center">Kakugo</div>
+<div align="center">
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/64b63f8ad57e02621dc93c8b/hmRaNkmPAV8rakBOhtgZI.png" alt="Globe Image" width="400"/>
+  A data distilled model trained specifically for <strong>Guarani</strong>.
+</div>
+</div>
+This is **Kakugo 3B Guarani**, a small language model (SLM) fine-tuned to interact with the user in **Guarani**.
+For Kakugo in other languages, check out the [model](https://huggingface.co/collections/ptrdvn/kakugo-models) and [dataset](https://huggingface.co/collections/ptrdvn/kakugo-datasets) collections.
+# How to use
+To use this model, you can use your preferred LLM inference package.
+This model should work with any package that supports the original base model [ibm-granite/granite-4.0-micro](https://huggingface.co/ibm-granite/granite-4.0-micro).
+We provide examples for how to run this with Huggingface or vLLM:
+<details>
+<summary>Huggingface (Recommended for beginners)</summary>
+First, make sure `transformers` is installed on your machine.
+```bash
+pip install transformers
+```
+Then run the following Python code to generate a response from the LLM.
+```python
+from transformers import pipeline
+generator = pipeline(model="ptrdvn/kakugo-3B-grn", task="text-generation")
+user_input = input("Please enter your input to the model in Guarani:")
+do_reasoning = False
+open_thinking_tag = "<think>"
+close_thinking_tag = "</think>"
+if do_reasoning:
+    sys_msg = f"Before you respond, first think about your response and enclose your thinking process in {open_thinking_tag} and {close_thinking_tag} delimiters."
+else:
+    sys_msg = "Be concise in your responses."
+message = [
+        {"role": "system", "content": sys_msg},
+        {"role": "user", "content": user_input}
+    ]
+output = generator(
+    message,
+    do_sample=False,
+    repetition_penalty=1.05,
+)
+model_response = output[0]["generated_text"][-1]["content"]
+if do_reasoning:
+    model_response = model_response.split(close_thinking_tag)[-1]
+print(model_response)
+```
+N.B. - We recommend using a `repetition_penalty` of 1.05 as sometimes the model can stuck in a loop of generating repetitive text when generating low-resource languages.
+You can set `do_reasoning` to be either True or False to turn "thinking mode" on or off, respectively. If the model is used in thinking mode, then it will take longer to generate a response, but may lead to a better generated response.
+This mode is still experimental, so try both using and not using it for your use-case.
+</details>
+<br/>
+<details>
+<summary>vLLM (Recommended for performance)</summary>
+First, make sure `vllm` is installed on your machine.
+```bash
+pip install vllm
+```
+Then run the following Python code to generate a response from the LLM.
+```python
+from vllm import LLM, SamplingParams
+llm = LLM(model="ptrdvn/kakugo-3B-grn")
+user_input = input("Please enter your input to the model in Guarani:")
+do_reasoning = True
+open_thinking_tag = "<think>"
+close_thinking_tag = "</think>"
+if do_reasoning:
+    sys_msg = f"Before you respond, first think about your response and enclose your thinking process in {open_thinking_tag} and {close_thinking_tag} delimiters."
+else:
+    sys_msg = "Be concise in your responses."
+sampling_params = SamplingParams(temperature=0, repetition_penalty=1.05, max_tokens=2048)
+messages = [[
+        {"role": "system", "content": sys_msg},
+        {"role": "user", "content": user_input}
+    ]]
+output = llm.chat(messages, sampling_params)
+model_response = output[0].outputs[0].text
+if do_reasoning:
+    model_response = model_response.split(close_thinking_tag)[-1]
+print(model_response)
+```
+N.B. - When using `vllm` for inference of multiple inputs, we recommend inputting them all at once. I.e., add more items to the outer list of the `messages` variable in the above script. [More on vLLM optimization](https://docs.vllm.ai/en/stable/configuration/optimization).
+We recommend using a `repetition_penalty` of 1.05 as sometimes the model can stuck in a loop of generating repetitive text when generating low-resource languages.
+You can set `do_reasoning` to be either True or False to turn "thinking mode" on or off, respectively. If the model is used in thinking mode, then it will take longer to generate a response, but may lead to a better generated response.
+This mode is still experimental, so try both using and not using it for your use-case.
+</details>
+<br/>
+# Training data
+The training data for this model can be found at [ptrdvn/kakugo-grn](https://huggingface.co/datasets/ptrdvn/kakugo-grn).
+This data was created by prompting [openai/gpt-oss-120b](https://huggingface.co/openai/gpt-oss-120b) to generate prompts and responses in Guarani.
+We also translate a set of prompts and responses from the [BAAI/Infinity-Instruct](https://huggingface.co/datasets/BAAI/Infinity-Instruct) dataset.
+More details about exactly how we created our data can be found in [our paper](https://arxiv.org/abs/2601.14051).
+# Training
+Full details of how this model was created (and how you can train a model in your own chosen language) can be found on our [Github repo](https://github.com/Peter-Devine/kakugo).
+To make this model, we fine-tuned [ibm-granite/granite-4.0-micro](https://huggingface.co/ibm-granite/granite-4.0-micro) for 1 epoch on [ptrdvn/kakugo-grn](https://huggingface.co/datasets/ptrdvn/kakugo-grn) using [Llama Factory](https://github.com/hiyouga/LlamaFactory).
+<details>
+<summary>Full Llama Factory training hyperparameters</summary>
+```yaml
+### model
+model_name_or_path: ibm-granite/granite-4.0-micro
+trust_remote_code: true
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+deepspeed: examples/deepspeed/ds_z3_config.json  # choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json]
+### dataset
+dataset_dir: /workspace/train
+dataset: ptrdvn/kakugo-grn
+template: granite4
+cutoff_len: 8000
+overwrite_cache: true
+preprocessing_num_workers: 16
+dataloader_num_workers: 4
+packing: true
+### Reporting
+report_to: wandb
+run_name: ptrdvn/kakugo-grn
+logging_steps: 1
+### output
+output_dir: ptrdvn/kakugo-grn
+save_strategy: "no"
+save_steps: 99999999
+plot_loss: true
+overwrite_output_dir: true
+save_only_model: true
+### train
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 1
+learning_rate: 1.0e-5
+num_train_epochs: 1.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.05
+bf16: true
+ddp_timeout: 180000000
+resume_from_checkpoint: null
+## eval
+val_size: 0.02
+per_device_eval_batch_size: 1
+eval_strategy: steps
+eval_steps: 0.2
+```
+</details>
+<br/>
+# Credits
+This model was trained by [@ptrdvn](https://huggingface.co/ptrdvn)
+If you use this model, please cite:
+```bibtex
+@article{devine2026kakugo,
+  title={Kakugo: Distillation of Low-Resource Languages into Small Language Models},
+  author={Devine, Peter and Sanni, Mardhiyah and Adilazuarda, Farid and Loizaga, Julieta Gil and Haddow, Barry},
+  journal={arXiv preprint arXiv:2601.14051},
+  year={2026}
+}
+```