100-1eb-pag-llama-3.1: Pangasinan LLaMA 3.1 8B (100% Checkpoint)

100-1eb-pag-llama-3.1 is the completed, 100% fine-tuned instruction-following language model tailored for Pangasinan (Pangasinense) and English-Pangasinan bilingual comprehension and generation.

It is fine-tuned from Meta-Llama-3.1-8B-Instruct over a full epoch on a curated Pangasinan dataset using Unsloth and exported to near-lossless 8-bit integer quantization (Q8_0).

---

##Model Highlights

• Base Architecture: Meta LLaMA 3.1 8B Instruct (RoPE 128k, GQA)
• Target Languages: Pangasinan (pag) & English (en)
• Checkpoint: 100% (Complete fine-tuning epoch, zero contamination)
• Format: GGUF Q8_0 (100-1eb-pag-llama-3.1-Q8_0.gguf ~8.5 GB) — Maximum precision 8-bit quantization for near-lossless generation
• Fine-Tuning Framework: Unsloth LoRA (r=16, alpha=16, target modules: q, k, v, o, gate, up, down)

---

Quickstart & Inference

1. Using llama.cpp / llama-cpp-python

pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="KiruuPixel/100-1eb-pag-llama-3.1",
    filename="100-1eb-pag-llama-3.1-Q8_0.gguf",
    n_gpu_layers=-1,  # Offload all layers to GPU
    n_ctx=2048,
)

prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
You are a helpful, respectful, and fluent multilingual AI assistant specialized in English and Pangasinan.

### Input:
Salaysayen mo no akin ya importante so edukasyon.

### Response:
"""

output = llm(prompt, max_tokens=256, temperature=0.3)
print(output["choices"][0]["text"])

2. Using Ollama

Download the model file and create a Modelfile:

FROM ./100-1eb-pag-llama-3.1-Q8_0.gguf
TEMPLATE """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{{ if .System }}{{ .System }}{{ else }}You are a helpful, respectful, and fluent multilingual AI assistant specialized in English and Pangasinan.{{ end }}

### Input:
{{ .Prompt }}

### Response:
"""
PARAMETER temperature 0.3
PARAMETER top_p 0.9
PARAMETER stop "<|eot_id|>"
PARAMETER stop "### Instruction:"

Create and run in terminal:

ollama create 100-1eb-pag-llama -f ./Modelfile
ollama run 100-1eb-pag-llama "Antoy sankarakepan ya pasyaran ed Pangasinan?"

---

🏛️ Dataset & Prompt Format

Fine-tuned using the standard Alpaca instruction template:

Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{instruction}

### Input:
{input}

### Response:
{response}

---

📄 Citation & Attribution

@misc{tawir2026pangasinan,
  title={100-1eb-pag-llama-3.1: 100% Instruction Fine-Tuned LLaMA 3.1 8B Q8 GGUF for Pangasinan},
  author={KiruuPixel},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/KiruuPixel/100-1eb-pag-llama-3.1}
}