timothydillan/gemma4-e2b-balinese-assistant-v8

Gemma 4 E2B Balinese Assistant v8 LoRA

Research LoRA adapter for timothydillan/gemma4-e2b-balinese-cpt, trained to improve lightweight Balinese and Indonesian assistant behavior.

This is not a merged model. Load it as a PEFT adapter on top of timothydillan/gemma4-e2b-balinese-cpt.

Intended Use

• Private research and evaluation for Balinese / Indonesian conversational assistants.
• Lightweight instruction-following experiments for regional-language model catalog work.
• Further supervised fine-tuning, preference tuning, or native-speaker review workflows.

Do not treat this checkpoint as production-ready, medically/legal/religiously authoritative, or native-speaker certified. Curated examples were written for repair and coverage, but remain pending native review.

Training Summary

• Base model: timothydillan/gemma4-e2b-balinese-cpt
• Training job: Kaggle kernel timothydillan/oim-balinese-gemma4-e2b-v0, version 15
• Dataset: data/processed/llm/sft_train_assistant_v8.jsonl
• Examples: 2,625
• Context length: 1,024
• Epochs: 2.0
• Steps: 658
• Learning rate: 2.5e-5
• LoRA rank / alpha: 16 / 16
• Batch / gradient accumulation: 2 / 4
• Final training loss: 0.4553
• Training runtime: about 2,165 seconds

Data Mix

Source	Examples
gemini-3.1-flash-lite-grounded-balinese	650
oim-curated-v7-behavior-repair-pending-native-review	317
indonlp/NusaX-MT	360
oim-curated-direct-v6-pending-native-review	343
akoksal/muri-it-language-split-ban	140
biznetgio/alpaca-clean-balinese	260
oim-curated-v8-dialogue-factuality-repair-pending-native-review	174
biznetgio/oasst2-balinese	180
ChavyvAkvar/aya_collection_language_split-balinese-Converted	120
pradanaadn/IndonesianNMT-balinese-refined	60
deepseek-grounded-balinese	10
deepseek-v4-flash-grounded-balinese	10
nusa-dialogue-ban-summ	1

v8 specifically adds dialogue and factuality repair rows after v7 fixed identity behavior but still failed a sampled dialogue case and one Bali geography fact. The new rows target complete A/B dialogue format, Bali/Denpasar factuality, Indonesian answer routing, uncertainty handling, concise answers, refusal boundaries, and reduction of English gloss artifacts.

Evaluation Status

This upload is the trained adapter. Hardened smoke evaluation is being tracked in the Open Indonesia Models repository and should be treated as the source of truth for current quality claims.

Expected checks include:

• identity stability
• Balinese food / cultural prompt handling
• canang sari explanation
• short dialogue completeness
• Bali / Denpasar factuality
• Indonesian language-switch behavior
• uncertainty and refusal handling
• loop / forbidden-phrase / malformed-salutation checks

Known Limits

• Low-resource Balinese coverage remains narrow.
• Native-speaker review has not been completed.
• The model may still hallucinate, mix registers, or produce awkward Balinese.
• Evaluation is smoke-test level unless a downstream report says otherwise.
• Dataset licenses and provenance should be reviewed before any public/commercial release beyond research experiments.

Loading

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoProcessor

base_id = "timothydillan/gemma4-e2b-balinese-cpt"
adapter_id = "timothydillan/gemma4-e2b-balinese-assistant-v8"

processor = AutoProcessor.from_pretrained(adapter_id)
base = AutoModelForCausalLM.from_pretrained(base_id, device_map="auto")
model = PeftModel.from_pretrained(base, adapter_id)

Use the Gemma chat template and keep generation conservative while evaluating.

Citation

If you use this checkpoint, cite the Open Indonesia Models project and the upstream base/model/data sources where applicable.