Add model card

README.md

@@ -1,21 +1,110 @@
 ---
-base_model: unsloth/meta-llama-3.1-8b-instruct-bnb-4bit
-tags:
-- text-generation-inference
-- transformers
-- unsloth
-- llama
-license: apache-2.0
 language:
-- en
+  - id
+  - ms
+base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
+tags:
+  - llama-3
+  - maluku
+  - melayu-maluku-utara
+  - fine-tuned
+  - unsloth
+  - lora
+  - gguf
+  - indonesian
+license: llama3
+pipeline_tag: text-generation
 ---
 
-# Uploaded finetuned  model
+# 🌴 Utu — Asisten Melayu Maluku Utara (Llama-3.1-8B)
+
+Model fine-tuned dari `meta-llama/Meta-Llama-3.1-8B-Instruct` untuk memahami dan merespons
+dalam **Bahasa Melayu Maluku Utara** (Bahasa Pasar/Bahasa Ternate).
+
+Fine-tuned menggunakan **Unsloth** untuk efisiensi maksimal di GPU T4.
+
+## Kosakata Lokal yang Dipahami
+
+| Kata | Arti | Contoh |
+|------|------|--------|
+| ngana / nga | kamu | "Ngana mo pi mana?" |
+| kita | saya | "Kita tra tau" |
+| torang / tong | kami/kita semua | "Torang mo pigi pasar" |
+| dorang / dong | mereka | "Dorang su pulang" |
+| pigi | pergi | "Kita mo pigi" |
+| tra / tara | tidak | "Kita tra mau" |
+| su / so | sudah | "Kita su makan" |
+| mo | mau/akan | "Ngana mo makan?" |
+| deng | dengan | "Kita deng ngana" |
+| bolom | belum | "Kita bolom makan" |
+| kobong | kebun | "Pigi di kobong" |
+| foya | bohong | "Jang foya" |
+| pe / p | kepemilikan | "Ini kita p buku" |
+| bkiapa | kenapa | "Bkiapa ngana sedih?" |
+
+## Cara Penggunaan
+
+### Python (Transformers)
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_id  = "haidar038/utu-malut"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model     = AutoModelForCausalLM.from_pretrained(
+    model_id, torch_dtype=torch.float16, device_map="auto"
+)
+
+messages = [
+    {"role": "system",    "content": "Ngana adalah Utu, asisten AI dari Ternate."},
+    {"role": "user",      "content": "Ngana mau pigi mana sore ini?"}
+]
+inputs  = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(model.device)
+outputs = model.generate(inputs, max_new_tokens=200, temperature=0.7, do_sample=True)
+print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))
+```
+
+### GGUF (CPU, via llama-cpp-python)
+```python
+from llama_cpp import Llama
+
+llm = Llama.from_pretrained(
+    repo_id  = "haidar038/utu-malut-GGUF",
+    filename = "*q4_k_m*.gguf",
+    n_ctx    = 512,
+)
+output = llm.create_chat_completion(messages=[
+    {"role": "system", "content": "Ngana adalah Utu, asisten AI dari Ternate."},
+    {"role": "user",   "content": "Ngana mau pigi mana?"},
+])
+print(output["choices"][0]["message"]["content"])
+```
+
+## Detail Training
+
+| Parameter | Nilai |
+|-----------|-------|
+| Base Model | meta-llama/Meta-Llama-3.1-8B-Instruct |
+| Fine-tuning | QLoRA 4-bit (Unsloth) |
+| LoRA r / alpha | 16 / 32 |
+| Dataset | ~450 train samples |
+| Epochs | 3 |
+| Learning rate | 0.0002 |
+| Max seq length | 512 |
+| Eval loss | 1.1841 |
+| Perplexity | 3.27 |
+| Platform | Kaggle (T4 GPU) |
+
+## Deployment
+
+Tersedia sebagai HF Space: [haidar038/utu-malut-chat](https://huggingface.co/spaces/haidar038/utu-malut-chat)
+
+## Limitasi
 
-- **Developed by:** haidar038
-- **License:** apache-2.0
-- **Finetuned from model :** unsloth/meta-llama-3.1-8b-instruct-bnb-4bit
+- Dataset ~500 baris; belum mencakup semua variasi dialek
+- Untuk riset dan pengembangan NLP bahasa daerah
+- Verifikasi output sebelum penggunaan produksi
 
-This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
+## Kredit
 
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
+Fine-tuned dengan [Unsloth](https://github.com/unslothai/unsloth) + [TRL](https://github.com/huggingface/trl) di Kaggle.