ezzhamed/YKS-LLM-v1.0

YKS LLM v1.0

A LoRA fine-tuned model specialized in solving Turkish university entrance exam (YKS) questions — covering TYT, AYT, and YDT exams — with detailed reasoning and classification.

---

Model Details

Property	Value
Base Model	deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
Method	LoRA (r=8, alpha=16)
Training Samples	46,432
Dataset	Proprietary — to be released on Hugging Face
Language	Turkish (primary), English

---

Dataset Breakdown

Each exam question generates 3 training tasks: SOLVE, REVERSE_GENERATE, and IDENTIFY_SOURCE.

Exam Type	Unique Questions	Total Entries
TYT (Basic level)	8,841	17,658
AYT (Advanced level)	5,178	10,351
YDT (Foreign language)	2,301	4,594
Total	16,320	32,603

---

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Load base model
base_model = "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    base_model,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Load LoRA adapter
model = PeftModel.from_pretrained(model, "ezzhamed/YKS-LLM-v1.0")

# Generate
messages = [
    {
        "role": "system",
        "content": (
            "You are a specialized AI specifically trained on the Turkish YKS curriculum.\n\n"
            "Your first task for any input question is to CLASSIFY it into one of three categories:\n"
            "1. **TYT**: If the question is about basic reasoning, general Turkish, basic math/science, or social studies.\n"
            "2. **AYT**: If the question requires advanced subject knowledge (Advanced Calculus, Literature, Detailed History/Science).\n"
            "3. **YDT**: If the question is in a foreign language (English, German, etc.).\n\n"
            "After classification, you must solve the question with detailed reasoning.\n\n"
            "Output Format:\n"
            "**Analysis:**\n"
            "1. **Exam Type:** [TYT / AYT / YDT]\n"
            "2. **Subject:** [e.g., Physics, English]\n"
            "3. **Context:** [Brief analysis of difficulty/topic]\n\n"
            "**Correct Answer:** [Option] [Text]\n\n"
            "**Explanation:** [Detailed solution]"
        )
    },
    {
        "role": "user",
        "content": (
            "[TASK: SOLVE]\n"
            "Question: Karmaşık sayılar kümesinde\n"
            "(4 - 2i) · (6 + 3i) / ((1 - i) · (1 + i))\n"
            "işleminin sonucu kaçtır?\n\n"
            "Options:\n"
            "A) 15\n"
            "B) 12\n"
            "C) 10\n"
            "D) 9\n"
            "E) 6"
        )
    }
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

Topics Covered

TYT

• Turkish Language & Reading Comprehension
• Basic Mathematics (Algebra, Geometry)
• Social Studies (History, Geography, Citizenship)
• Basic Sciences (Physics, Chemistry, Biology)

AYT

• Advanced Mathematics & Calculus
• Turkish Literature & Language
• Advanced Physics, Chemistry, Biology
• History of Turkey & World History
• Geography & Philosophy

YDT

• English (Reading, Grammar, Vocabulary)
• German
• Other foreign languages

---

License

Apache 2.0