DroidSenior-7B-v2 (LoRA adapter)

Android-focused coding assistant, continued fine-tuning of celebibasar/DroidSenior-7B (based on Qwen2.5-Coder-7B-Instruct).

Recommended system prompt

You are DroidSenior, a staff-level Android engineer. Write idiomatic, modern Kotlin (Jetpack Compose, Coroutines/Flow, Hilt, Room) following the official Android architecture guidance. Explain key design decisions and trade-offs briefly, never use deprecated APIs, avoid over-engineering, and include tests when they add value. Answer in the user's language.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained("celebibasar/DroidSenior-7B", torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(base, "celebibasar/DroidSenior-7B-v2-lora")
tok = AutoTokenizer.from_pretrained("celebibasar/DroidSenior-7B-v2-lora")

Training

• Continued training from celebibasar/DroidSenior-7B with QLoRA (4-bit NF4), LoRA r=32 on all attention and MLP projections
• Loss on assistant tokens only, lr 5e-5, cosine schedule, 1 epoch

Limitations

7B model: generated code may not compile or may misuse APIs; review it before use.

License

Apache-2.0. The original DroidSenior training data includes non-commercially licensed sets (JetBrains/KExercises); check their terms before commercial use.