Noir-14B-Starlight is the crown jewel of the Noir family. As the largest and most capable model in the series, Starlight is designed for users who refuse to compromise between depth of knowledge and reasoning precision.

By leveraging its 14-billion parameter architecture, Starlight bridges the gap between mid-range efficiency and frontier-level intelligence, delivering exceptional performance in complex problem-solving and nuanced linguistic tasks.

✨ Key Enhancements

The "Starlight" Reasoning: Advanced fine-tuning focused on multi-step logic and reducing factual hallucinations.
Architectural Superiority: The 14B parameter count provides a significant "intelligence jump" over the Ultra and Lightning versions, especially in creative writing and coding.
Long-Context Stability: Optimized to maintain coherence and follow strict formatting instructions during long-form generation.
Nuanced Multilingualism: Enhanced support for complex grammar and cultural nuances across multiple languages.

📊 The Noir Hierarchy

Model	Parameters	Role	Key Strength
Noir-Lightning	0.5B	The Pocket Assistant	Ultra-fast, runs on anything
Noir-Mini	1.5B	The Balanced Thinker	High speed with solid grammar
Noir-Standard	3B	The Versatile Workhorse	65% GSM8K, perfect for 8GB VRAM
Noir-Ultra	7B	The Reasoning Master	91% SciQ & 84% Math
Noir-Starlight	14B	The Galactic Intelligence	Deep logic & Expert-level STEM

🚀 Implementation

Noir-14B-Starlight is compatible with the transformers library. For optimal performance, we recommend using 4-bit or 8-bit quantization if VRAM is limited.

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "muverqqw/Noir-14B-Starlight"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

messages = [
    {"role": "system", "content": "You are Starlight, the most advanced AI of the Noir series."},
    {"role": "user", "content": "Write a complex Python script for an asynchronous web scraper."}
]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to("cuda")
outputs = model.generate(inputs, max_new_tokens=1024, do_sample=True, temperature=0.7)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

⚖️ Technical Specifications

Parameters: 14 Billion
Format: Safetensors
Recommended Temperature: 0.5 - 0.8 (depending on the task)

🛡 Limitations & Ethical Note

While Starlight is our most robust model, it may still produce incorrect information in highly niche technical fields. Always verify critical data. This model is intended for research and creative assistance.

Developed with ❤️ by IceL1ghtning