Vorenthos-r1
Fine-tuned version of deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B using Unsloth + QLoRA.
Training details
| Setting | Value |
|---|---|
| Base model | deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B |
| Fine-tuning method | QLoRA (r=16, alpha=16) |
| Precision (saved) | float16 (merged) |
| Max sequence length | 2048 |
| Batch size (effective) | 8 |
| Epochs | 1 |
| Learning rate | 0.0002 |
| Date trained | 2026-07-26 |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_id = "vorenthiclabs/Vorenthos-r1"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float16,
device_map="auto",
)
prompt = """<|User|>: Explain gradient descent in simple terms.
<|Assistant|>: <think>
"""
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.6)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Prompt format
<|User|>: {your instruction here}
<|Assistant|>: <think>
{model chain-of-thought reasoning}
</think>
{final response}
License
MIT — see base model card for additional terms.