Fine-tuned Qwen3-14B Model

Model Description

This is a fine-tuned version of the Qwen3-14B model optimized with Unsloth and 4-bit quantization for efficient inference. The model was fine-tuned on a curated historical text corpus focused on early 20th-century diplomatic events and treaties, enabling nuanced understanding and generation of complex political and historical narratives.

Key Features

2x faster training using Unsloth optimization
4-bit quantization for reduced memory footprint and efficient inference
Enhanced understanding of historical and diplomatic discourse
Compatible with Hugging Face’s TRL library
Supports LoRA fine-tuning for flexible adaptation

Model Details

Base Model: unsloth/qwen3-14b-unsloth-bnb-4bit
License: Apache 2.0
Fine-tuned by: momererkoc
Languages: Primarily Turkish, with English capabilities
Quantization: 4-bit via BitsAndBytes

Training Data

Extensive historical text collection focusing on early 20th-century diplomacy and treaty analysis
Texts rich in complex political context and multi-layered narrative

Usage Example

from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "momererkoc/qwen3-14b-lozan", # YOUR MODEL YOU USED FOR TRAINING
    max_seq_length = 2048,
    load_in_4bit = True,
    token="hf_...",
)

messages = [
    {"role" : "user", "content" : "Lozan anlaşmasının sonuçları nelerdir?"}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize = False,
    add_generation_prompt = True, # Must add for generation
    enable_thinking = False, # Disable thinking
)

from transformers import TextStreamer
_ = model.generate(
    **tokenizer(text, return_tensors = "pt").to("cuda"),
    max_new_tokens = 256, # Increase for longer outputs!
    temperature = 0.7, top_p = 0.8, top_k = 20, # For non thinking
    streamer = TextStreamer(tokenizer, skip_prompt = True),
)

Performance

2x faster fine-tuning compared to standard methods
Reduced GPU memory usage due to 4-bit quantization
Strong contextual understanding of historical and political texts
Supports generation of coherent, context-aware historical narratives

Optimization Details

Uses Unsloth for accelerated training
4-bit quantization via BitsAndBytes
LoRA for parameter-efficient fine-tuning

Acknowledgments

Unsloth team for optimization tools
Hugging Face for the transformers ecosystem
Dataset creators for curated historical texts

momererkoc
/

qwen3-14b-lozan