teolm30/fox1.4-high-reasoning

Fox 1.4 High Reasoning

A Greek fine-tuned version of Fox 1.3, a Qwen2.5-1B-Instruct based model.

Training

• Method: QLoRA (4-bit quantization)
• Dataset: Custom Greek conversation dataset (20 conversations)
• Epochs: 3
• Loss: 3.26 → 0.79
• Hardware: NVIDIA RTX 3050 (6GB VRAM)
• LoRA rank: 8, alpha: 16

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("teolm30/fox1.4-high-reasoning", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("teolm30/fox1.4-high-reasoning", trust_remote_code=True)

messages = [{"role": "user", "content": "Τι είναι η τεχνητή νοημοσύνη;"}]
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt")
output = model.generate(input_ids, max_new_tokens=256)
print(tokenizer.decode(output[0]))

Greek Language Support

Fine-tuned for Greek language — grammar, vocabulary, natural conversation, and technical topics in Greek.

Previous Versions

• Fox 1.3 — base model
• Fox 1.0 — original model