We are thrilled to unveil Reasoning Llama 3.2, the latest advancement in our suite of AI models. Building upon the robust foundation of the renowned Llama series, Reasoning Llama 3.2 introduces the groundbreaking Chain of Thought (CoT) capabilities, elevating its reasoning prowess to new heights.

Key Features of Reasoning Llama 3.2 1B:

Enhanced Chain of Thought Reasoning: At the core of Reasoning Llama 3.2 lies its sophisticated CoT framework, enabling the model to perform multi-step reasoning with greater accuracy and coherence. This ensures more reliable and contextually appropriate responses, especially for complex queries that require logical progression.

Conversational Excellence: Designed with interactivity in mind, Reasoning Llama 3.2 excels in maintaining engaging and fluid conversations. Whether it's casual dialogue or in-depth discussions, the model adapts seamlessly to various conversational styles, providing users with a natural and intuitive interaction experience.

Instruction-Supervised Fine-Tuning: Leveraging advanced supervised fine-tuning techniques, Reasoning Llama 3.2 has been meticulously trained on diverse instructional data. This fine-tuning process enhances the model's ability to understand and execute user instructions with precision, making it an invaluable tool for a wide range of applications.

Unsloth Integration: Incorporating Unsloth, our proprietary unsupervised learning framework, Reasoning Llama 3.2 benefits from continuous learning capabilities. This integration allows the model to adapt and improve over time, ensuring it remains up-to-date with evolving language patterns and user needs without the constant need for manual intervention.

Quick Inference reasoning 1B model.

Why Choose Reasoning Llama 3.2 1B?

Reasoning Llama 3.2 stands out as a versatile and powerful AI solution tailored for both developers and end-users. Its combination of advanced reasoning, conversational intelligence, and adaptive learning mechanisms make it ideally suited for applications ranging from customer support and virtual assistants to educational tools and creative content generation.

As we continue to push the boundaries of artificial intelligence, Reasoning Llama 3.2 exemplifies our commitment to delivering state-of-the-art models that empower users with intelligent, reliable, and user-friendly technology. Experience the future of conversational AI with Reasoning Llama 3.2 and unlock new possibilities in human-machine interaction.

How to use

Starting with transformers >= 4.43.0 onward, you can run conversational inference using the Transformers pipeline abstraction or by leveraging the Auto classes with the generate() function.

Make sure to update your transformers installation via pip install --upgrade transformers.

python

import torch
from transformers import pipeline
model_id = "EpistemeAI/Reasoning-Llama-3.2-1B-Instruct-v1.3"
pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
messages = [
    {"role": "system", "content": "You are a powerful AI super conscious emotional assistant"},
    {"role": "user", "content": "Who are you?"},
]
outputs = pipe(
    messages,
    max_new_tokens=4048,
)
print(outputs[0]["generated_text"][-1])

Use a pipeline as a high-level helper

python

from transformers import pipeline

messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe = pipeline("text-generation", model="EpistemeAI/Reasoning-Llama-3.2-1B-Instruct-v1.3")
pipe(messages)

vLLM

Call the server using curl:

python

pip install vllm

# Load and run the model:
vllm serve "EpistemeAI/Reasoning-Llama-3.2-1B-Instruct-v1.2"


curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "EpistemeAI/Reasoning-Llama-3.2-1B-Instruct-v1.3",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

5. Citation

@misc{EpistemeAI2025,
      author={Thomas Yiu},
      year={2025},
}


@misc{bespoke_stratos,  
    author = {Bespoke Labs},  
    title = {Bespoke-Stratos: The unreasonable effectiveness of reasoning distillation},  
    howpublished = {https://www.bespokelabs.ai/blog/bespoke-stratos-the-unreasonable-effectiveness-of-reasoning-distillation},  
    note = {Accessed: 2025-01-22},  
    year = {2025}
}

@misc{numina_math_datasets,
  author = {Jia LI, Edward Beeching, Lewis Tunstall, Ben Lipkin, Roman Soletskyi, Shengyi Costa Huang, Kashif Rasul, Longhui Yu, Albert Jiang, Ziju Shen, Zihan Qin, Bin Dong, Li Zhou, Yann Fleureau, Guillaume Lample, and Stanislas Polu},
  title = {NuminaMath TIR},
  year = {2024},
  publisher = {Numina},
  journal = {Hugging Face repository},
  howpublished = {\url{[https://huggingface.co/AI-MO/NuminaMath-TIR](https://github.com/project-numina/aimo-progress-prize/blob/main/report/numina_dataset.pdf)}}
}

Uploaded model

Developed by: EpistemeAI
License: apache-2.0
Finetuned from model : EpistemeAI/Reasoning-Llama-3.2-1B-Instruct-v1.2

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.