Lumma-0.6B-Instruct

Lumma-0.6B-Instruct

Introduction

Lumma-0.6B-Instruct is a compact, efficient multilingual language model designed for strong performance in resource-constrained environments. It is pre-trained from scratch on 1 trillion tokens and further enhanced through instruction tuning and Direct Preference Optimisation. This is a pre-RL checkpoint. The model supports English and 10 Indic languages.

Benchmark results

<img src="benchmark_image.jpg" width="1500"/>

We benchmarked Lumma-0.6B-Instruct across multiple benchmarks, with an intentional focus on instruction-following capabilities. Despite its compact size, Lumma-0.6B-Instruct is able to match or outperform similar models up to 3× larger on several instruction-following benchmarks.

While the model also delivers decent performance on mathematics and coding, we believe these capabilities are less critical for the primary real-world use cases targeted by such a small model, where developers typically prioritize efficient and reliable instruction following.

We expect further improvements with the RL-trained version of Lumma-0.6B-Instruct, particularly as we continue optimizing the model for real-world instruction-following tasks.

🌍 Supported Languages

The model is trained on English and a diverse set of Indic languages, including Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Odia

🚀 Usage

!pip install transformers=='5.4.0'

from IPython.display import display, Markdown
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "FrontiersMind/Lumma-0.6B-Instruct"

device = "cuda" if torch.cuda.is_available() else "cpu"

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    trust_remote_code=True,
    dtype=torch.bfloat16
).to(device).eval()

prompt = "Explain newton's second law of motion"

messages = [
    {"role": "user", "content": prompt}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **inputs,
    max_new_tokens=500,
    do_sample=True,
    temperature=0.3,
    top_p=0.90,
    top_k=20,
    repetition_penalty=1.1,
)

generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

#print(response)
Markdown(response)

📬 Feedback & Suggestions

We’d love to hear your thoughts, feedback, and ideas!

• Discord: https://discord.gg/ZGdjCdRt
• Email: [email protected]
• Official Website https://www.frontiersmind.ai/
• LinkedIn: https://www.linkedin.com/company/frontiersmind/
• X (Twitter): https://x.com/FrontiersMind