KrynexAI 1.4

KrynexAI 1.4 is a compact, ultra-lightweight 150M-parameter language model designed for high-efficiency text generation, fast inference, and on-device deployment.

🧠 Model Summary

KrynexAI 1.4 features an optimized Llama-based architecture trained on a multi-trillion token mix of web content, code, and educational data. It provides strong base capabilities in reasoning, instruction following, and general text processing while maintaining a minimal memory footprint (~270 MB weights).

• Architecture: Llama-based Transformer Decoder
• Parameters: ~150M
• Context Window: 8,192 tokens
• Precision: bfloat16 / fp16

🚀 Quickstart & Usage

pip install transformers

🐍 Basic Inference (Python)

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
checkpoint = "KrynexLabs/KrynexAI-1.4-150M-Mobile-Model"
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.from_pretrained(
    checkpoint, 
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
inputs = tokenizer.encode("Gravity is", return_tensors="pt").to(device)
outputs = model.generate(inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📌 Performance Benchmarks

Warning: This repository may contain inaccurate, outdated, or otherwise unsuitable files. Approximately half of the original files were left unchanged, so their contents may not reflect the current state of the project or may no longer be relevant.

🎥 Full Video Review: Tests & Failures

Watch a full 44-minute independent review covering generation tests, prompt handling, and honest breakdowns of where this 150M model succeeds and fails.

Note: As a small language model (SLM), it has limitations, and the video shows both its strengths and its hallucinations.

🎬 Watch the full review on YouTube

Click the image above to watch on YouTube.

📜 License

This model is released under Apache-2.0 License — you can use, modify, and share this code freely, even in commercial projects. The only requirement is to keep the original attribution. Full text in the file: LICENSE.