LLM2026_DPO_SFT19_v1

This model is a fine-tuned version of makotonlo/LLM2026_SFT_finalv19 using Direct Preference Optimization (DPO) via the Unsloth library.

This repository contains the full-merged 16-bit weights. No adapter loading is required.

Training Objective

This model has been optimized using DPO to eliminate unwanted formatting (like markdown code blocks and introductory chatter) and to ensure pure structured data output.

Training Configuration

• Base model: makotonlo/LLM2026_SFT_finalv19
• Method: DPO (Direct Preference Optimization)
• Epochs: 1
• Learning rate: 1e-06
• Beta: 0.1
• Max sequence length: 1024
• LoRA Config: r=64, alpha=64 (merged into base)

Usage

Since this is a merged model, you can use it directly with transformers.

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "makotonlo/LLM2026_DPO_SFT19_v1" # 自動反映

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Test inference
prompt = "Your question here"
inputs = tokenizer.apply_chat_template([{ "role": "user", "content": prompt }], tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))