ozkurt7/oracle-deepseek-r1-adapter

Oracle Fusion Cloud SCM - DoRA Adapter

Bu Oracle Fusion Cloud SCM konularında uzmanlaşmış DoRA (Weight-Decomposed Low-Rank Adaptation) adapter'ıdır.

🎯 Kullanım

Google Colab'da Merge:

# 1. Base model yükle
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

base_model_name = "unsloth/DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit"
adapter_name = "ozkurt7/oracle-deepseek-r1-adapter"

# 2. Model ve adapter yükle  
tokenizer = AutoTokenizer.from_pretrained(base_model_name)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)

model = PeftModel.from_pretrained(base_model, adapter_name)

# 3. Merge işlemi
merged_model = model.merge_and_unload()

# 4. Kaydet
merged_model.save_pretrained("./oracle-merged")
tokenizer.save_pretrained("./oracle-merged")

# 5. Test
messages = [
    {"role": "system", "content": "You are an Oracle Fusion Cloud SCM expert."},
    {"role": "user", "content": "What is Oracle SCM Cloud?"}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
outputs = merged_model.generate(**inputs, max_new_tokens=200, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Local kullanım:

# Adapter'ı indir
git clone https://huggingface.co/ozkurt7/oracle-deepseek-r1-adapter

# Python'da merge
python merge_adapter.py

📊 Model Details

• Base Model: unsloth/DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit
• Technique: DoRA (Weight-Decomposed Low-Rank Adaptation)
• Domain: Oracle Fusion Cloud SCM
• Status: Adapter only (merge required)
• Memory: ~500MB (adapter only)

🚀 Next Steps

1. Google Colab'da bu adapter'ı kullanarak merge yapın
2. Merge edilen modeli yeni repo'ya upload edin
3. GGUF formatına dönüştürün

🛠️ Troubleshooting

• Memory Error: Colab Pro kullanın veya local'de merge yapın
• Loading Error: trust_remote_code=True ekleyin
• CUDA Error: device_map="auto" kullanın

Created by: Kaggle → Google Colab workflow Date: 2025-08-12