luigicfilho/Numera-v1-3b

Numera: The Numerically Generated Model

Numera-v1 is a 3 Billion parameter transformer-based causal decoder model designed for high-quality text generation and conceptual coherence. It is part of the Numera series focusing on structural integrity and advanced architectural optimization.

Model Details

• Model Name: Numera (Gen-1) 3B
• Generated By: LCDev-Numera
• Base Architecture: Qwen2.5
• Parameters: 3 Billion
• Type: Statistical Weight Generation
• Date Generated: 2026-02-20

Model Technical Specifications

Here are the details for Numera (Gen-1) 3b :

• Total Parameters: ~3.09 Billion
• Architecture: Qwen2.5 Family
  • Layers: 36
  • Attention Heads: 16 (Query) / 2 (KV - Grouped Query Attention)
  • Hidden Size: 2048
  • Intermediate Size (MLP): 11008
• Vocab Size: 151,936 tokens
• Context Window: 32,768 tokens
• Format: SafeTensors (Universal, safe serialization)
• Nature: Numerically Generated (Non-trained, statistical approximation)

Intended Use

This model is intended for research into:

• Weight space analysis of Large Language Models.
• Statistical properties of model weights.
• Experimental initialization checkpoints.

Note: This model is a statistical approximation and not a trained model. It may exhibit repetitive behaviors or lack specific factual knowledge.

Usage

You can use this model with the transformers library:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "luigicfilho/Numera-v1-3b"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

prompt = "The nature of mathematics is"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

License

This model is released under the MIT License.