---
language:
- en
- ko
license: other
license_name: exaone
license_link: LICENSE
tags:
- text-generation-inference
- transformers
- trl
- sft
- reasoning
- lg-ai
- exaone
- exaone-3.5
- o1
base_model: LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct
datasets:
- KingNish/reasoning-base-20k
---

# Model Description

An uncensored reasoning EXAONE 3.5 model trained on reasoning data. Now with a full epoch!

It has been trained using improved training code, and gives an improved performance.
Here is what inference code you should use:
```py
from transformers import AutoModelForCausalLM, AutoTokenizer

MAX_REASONING_TOKENS = 1024
MAX_RESPONSE_TOKENS = 512

model_name = "lunahr/thea-pro-2b-100r"

model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "Which is greater 9.9 or 9.11 ??"
messages = [
    {"role": "user", "content": prompt}
]

# Generate reasoning
input_ids = tokenizer.apply_chat_template(messages, tokenize=False, add_reasoning_prompt=True, return_tensors="pt")
output = model.generate(
    input_ids.to("cuda"),
    eos_token_id=tokenizer.eos_token_id,
    max_new_tokens=MAX_REASONING_TOKENS,
    do_sample=False,
)

print("REASONING: " + tokenizer.decode(output[0]))

# Generate answer
messages.append({"role": "reasoning", "content": reasoning_output})
input_ids = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, return_tensors="pt")
output = model.generate(
    input_ids.to("cuda"),
    eos_token_id=tokenizer.eos_token_id,
    max_new_tokens=MAX_RESPONSE_TOKENS,
    do_sample=False,
)

print("REASONING: " + tokenizer.decode(output[0]))
```

- **Trained by:** [Piotr Zalewski](https://huggingface.co/lunahr)
- **License:** exaone
- **Finetuned from model:** [LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct](https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct)
- **Dataset used:** [KingNish/reasoning-base-20k](https://huggingface.co/datasets/KingNish/reasoning-base-20k)

This Llama model was trained faster than [Unsloth](https://github.com/unslothai/unsloth) using [custom training code](https://www.kaggle.com/code/piotr25691/distributed-hf-training-with-2xt4).

Visit https://www.kaggle.com/code/piotr25691/distributed-hf-training-with-2xt4 to find out how you can finetune your models using BOTH of the Kaggle provided GPUs.