You can deploy this model and start earning money today!

efedemircan/
qwen3-14b-turkish-finance

Name: qwen3-14b-turkish-finance
Availability: InStock

transformers

safetensors

text-generation-inference

unsloth

qwen3

trl

AlicanKiraz0/Turkish-Finance-SFT-Dataset

apache-2.0

endpoints_compatible

Views

No views yet

Qwen3-14B Turkish Finance

A fine-tuned version of Qwen3-14B specialized in Turkish and global financial markets, including cryptocurrency, stock markets, technical analysis, fundamental analysis, and risk management — with a focus on the Turkish investor perspective.

Model Details

Property	Value
Base Model	Qwen3-14B
Fine-tuning Method	LoRA (Low-Rank Adaptation)
Training Framework	Unsloth + TRL (SFTTrainer)
Quantization	4-bit (bnb)
LoRA Rank	32
LoRA Alpha	32
Max Sequence Length	8192
Training Steps	600 (2 epochs)
Final Training Loss	0.917
Language	Turkish
Domain	Finance (Crypto, BIST, NASDAQ, Technical Analysis, Risk Management)

Intended Use

This model is designed for:

Answering questions about cryptocurrency ecosystems (Bitcoin, Ethereum, DeFi, Layer 2, stablecoins)
Providing insights on Turkish stock markets (BIST 100, BIST 30, sector indices)
Discussing global markets (NASDAQ, S&P 500, DAX, Nikkei)
Explaining technical analysis concepts (chart patterns, indicators, Fibonacci, Elliott Wave)
Explaining fundamental analysis (DCF, P/E, EV/EBITDA, tokenomics)
Guiding on risk management (portfolio theory, position sizing, hedging)
Explaining regulations relevant to Turkish and global investors (SPK, SEC, MiCA)

⚠️ Disclaimer: This model is intended for educational and informational purposes only. It does not provide investment advice. Always consult a licensed financial advisor before making investment decisions.

How to Use

Load the Model

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name     = "efedemircan/qwen3-14b-turkish-finance",
    max_seq_length = 8192,
    load_in_4bit   = True,
)
FastLanguageModel.for_inference(model)

Inference (Recommended — Non-Thinking Mode)

python

messages = [
    {
        "role": "user",
        "content": "Türkiye'de enflasyonist ortamlarda yatırımcılar hangi varlık sınıflarına yönelir?"
    }
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize              = False,
    add_generation_prompt = True,
    enable_thinking       = False,  # Fast, direct responses
)

inputs = tokenizer(text, return_tensors="pt").to("cuda")

outputs = model.generate(
    **inputs,
    max_new_tokens     = 2048,
    temperature        = 0.7,
    top_p              = 0.8,
    top_k              = 20,
    repetition_penalty = 1.1,
)

response = tokenizer.decode(
    outputs[0][inputs["input_ids"].shape[1]:],
    skip_special_tokens=True
)
print(response)

Inference (Thinking Mode)

For more complex analytical questions, enable the thinking mode to allow chain-of-thought reasoning:

python

text = tokenizer.apply_chat_template(
    messages,
    tokenize              = False,
    add_generation_prompt = True,
    enable_thinking       = True,   # Enables <think> block
)

outputs = model.generate(
    **inputs,
    max_new_tokens     = 8192,
    temperature        = 0.6,
    top_p              = 0.95,
    top_k              = 20,
    repetition_penalty = 1.15,
)

Training Details

Dataset

The model was fine-tuned on the Turkish Finance SFT Dataset, which contains 2,397 high-quality Turkish finance Q&A pairs covering:

Cryptocurrency ecosystems and on-chain analysis
BIST and global stock markets
Technical and fundamental analysis
Risk management strategies
Turkish and global financial regulations

Limitations

The model may occasionally produce hallucinated financial data (e.g., incorrect ticker codes, non-existent products). Always verify specific figures and product names from official sources.
The training dataset contains 2,397 examples, which is relatively small. Performance on highly niche or rarely discussed topics may be limited.
The model reflects knowledge present in the training dataset and the base Qwen3-14B model. It may not reflect real-time market data or recent regulatory changes.
Not suitable for automated trading systems or any application where financial accuracy is mission-critical.

Citation

If you use this model, please cite both the base model and the training dataset:

bibtex

@dataset{kiraz2025turkishfinance,
  title     = {Turkish Finance SFT Dataset},
  author    = {Kiraz, Alican},
  year      = {2026},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/AlicanKiraz0/Turkish-Finance-SFT-Dataset}
}

@misc{qwen3-14b-turkish-finance,
  title  = {Qwen3-14B Turkish Finance},
  author = {Demircan, Efe},
  year   = {2026},
  url    = {https://huggingface.co/efedemircan/qwen3-14b-turkish-finance}
}

Acknowledgements

Unsloth for the efficient fine-tuning framework
Qwen Team for the base Qwen3-14B model
Alican Kiraz for the Turkish Finance SFT Dataset

Uploaded model

Developed by: efedemircan
License: apache-2.0
Finetuned from model : unsloth/qwen3-14b-unsloth-bnb-4bit

This qwen3 model was trained 2x faster with Unsloth