Mungert/X-Ray_Alpha-GGUF

X-Ray_Alpha GGUF Models

Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)

Our latest quantization method introduces precision-adaptive quantization for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on Llama-3-8B. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.

Benchmark Context

All tests conducted on Llama-3-8B-Instruct using:

• Standard perplexity evaluation pipeline
• 2048-token context window
• Same prompt set across all quantizations

Method

• Dynamic Precision Allocation:
  • First/Last 25% of layers → IQ4_XS (selected layers)
  • Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
• Critical Component Protection:
  • Embeddings/output layers use Q5_K
  • Reduces error propagation by 38% vs standard 1-2bit

Quantization Performance Comparison (Llama-3-8B)

Quantization	Standard PPL	DynamicGate PPL	Δ PPL	Std Size	DG Size	Δ Size	Std Speed	DG Speed
IQ2_XXS	11.30	9.84	-12.9%	2.5G	2.6G	+0.1G	234s	246s
IQ2_XS	11.72	11.63	-0.8%	2.7G	2.8G	+0.1G	242s	246s
IQ2_S	14.31	9.02	-36.9%	2.7G	2.9G	+0.2G	238s	244s
IQ1_M	27.46	15.41	-43.9%	2.2G	2.5G	+0.3G	206s	212s
IQ1_S	53.07	32.00	-39.7%	2.1G	2.4G	+0.3G	184s	209s

Key:

• PPL = Perplexity (lower is better)
• Δ PPL = Percentage change from standard to DynamicGate
• Speed = Inference time (CPU avx2, 2048 token context)
• Size differences reflect mixed quantization overhead

Key Improvements:

• 🔥 IQ1_M shows massive 43.9% perplexity reduction (27.46 → 15.41)
• 🚀 IQ2_S cuts perplexity by 36.9% while adding only 0.2GB
• ⚡ IQ1_S maintains 39.7% better accuracy despite 1-bit quantization

Tradeoffs:

• All variants have modest size increases (0.1-0.3GB)
• Inference speeds remain comparable (<5% difference)

When to Use These Models

📌 Fitting models into GPU VRAM

✔ Memory-constrained deployments

✔ Cpu and Edge Devices where 1-2bit errors can be tolerated

✔ Research into ultra-low-bit quantization

Choosing the Right Model Format

Selecting the correct model format depends on your hardware capabilities and memory constraints.

BF16 (Brain Float 16) – Use if BF16 acceleration is available

• A 16-bit floating-point format designed for faster computation while retaining good precision.
• Provides similar dynamic range as FP32 but with lower memory usage.
• Recommended if your hardware supports BF16 acceleration (check your device's specs).
• Ideal for high-performance inference with reduced memory footprint compared to FP32.

📌 Use BF16 if:
✔ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
✔ You want higher precision while saving memory.
✔ You plan to requantize the model into another format.

📌 Avoid BF16 if:
❌ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.

---

F16 (Float 16) – More widely supported than BF16

• A 16-bit floating-point high precision but with less of range of values than BF16.
• Works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
• Slightly lower numerical precision than BF16 but generally sufficient for inference.

📌 Use F16 if:
✔ Your hardware supports FP16 but not BF16.
✔ You need a balance between speed, memory usage, and accuracy.
✔ You are running on a GPU or another device optimized for FP16 computations.

📌 Avoid F16 if:
❌ Your device lacks native FP16 support (it may run slower than expected).
❌ You have memory limitations.

---

Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference

Quantization reduces model size and memory usage while maintaining as much accuracy as possible.

• Lower-bit models (Q4_K) → Best for minimal memory usage, may have lower precision.
• Higher-bit models (Q6_K, Q8_0) → Better accuracy, requires more memory.

📌 Use Quantized Models if:
✔ You are running inference on a CPU and need an optimized model.
✔ Your device has low VRAM and cannot load full-precision models.
✔ You want to reduce memory footprint while keeping reasonable accuracy.

📌 Avoid Quantized Models if:
❌ You need maximum accuracy (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).

---

Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)

These models are optimized for extreme memory efficiency, making them ideal for low-power devices or large-scale deployments where memory is a critical constraint.

• IQ3_XS: Ultra-low-bit quantization (3-bit) with extreme memory efficiency.

  • Use case: Best for ultra-low-memory devices where even Q4_K is too large.
  • Trade-off: Lower accuracy compared to higher-bit quantizations.

• IQ3_S: Small block size for maximum memory efficiency.

  • Use case: Best for low-memory devices where IQ3_XS is too aggressive.

• IQ3_M: Medium block size for better accuracy than IQ3_S.

  • Use case: Suitable for low-memory devices where IQ3_S is too limiting.

• Q4_K: 4-bit quantization with block-wise optimization for better accuracy.

  • Use case: Best for low-memory devices where Q6_K is too large.

• Q4_0: Pure 4-bit quantization, optimized for ARM devices.

  • Use case: Best for ARM-based devices or low-memory environments.

---

Summary Table: Model Format Selection

Model Format	Precision	Memory Usage	Device Requirements	Best Use Case
BF16	Highest	High	BF16-supported GPU/CPUs	High-speed inference with reduced memory
F16	High	High	FP16-supported devices	GPU inference when BF16 isn't available
Q4_K	Medium Low	Low	CPU or Low-VRAM devices	Best for memory-constrained environments
Q6_K	Medium	Moderate	CPU with more memory	Better accuracy while still being quantized
Q8_0	High	Moderate	CPU or GPU with enough VRAM	Best accuracy among quantized models
IQ3_XS	Very Low	Very Low	Ultra-low-memory devices	Extreme memory efficiency and low accuracy
Q4_0	Low	Low	ARM or low-memory devices	llama.cpp can optimize for ARM devices

---

Included Files & Details

X-Ray_Alpha-bf16.gguf

• Model weights preserved in BF16.
• Use this if you want to requantize the model into a different format.
• Best if your device supports BF16 acceleration.

X-Ray_Alpha-f16.gguf

• Model weights stored in F16.
• Use if your device supports FP16, especially if BF16 is not available.

X-Ray_Alpha-bf16-q8_0.gguf

• Output & embeddings remain in BF16.
• All other layers quantized to Q8_0.
• Use if your device supports BF16 and you want a quantized version.

X-Ray_Alpha-f16-q8_0.gguf

• Output & embeddings remain in F16.
• All other layers quantized to Q8_0.

X-Ray_Alpha-q4_k.gguf

• Output & embeddings quantized to Q8_0.
• All other layers quantized to Q4_K.
• Good for CPU inference with limited memory.

X-Ray_Alpha-q4_k_s.gguf

• Smallest Q4_K variant, using less memory at the cost of accuracy.
• Best for very low-memory setups.

X-Ray_Alpha-q6_k.gguf

• Output & embeddings quantized to Q8_0.
• All other layers quantized to Q6_K .

X-Ray_Alpha-q8_0.gguf

• Fully Q8 quantized model for better accuracy.
• Requires more memory but offers higher precision.

X-Ray_Alpha-iq3_xs.gguf

• IQ3_XS quantization, optimized for extreme memory efficiency.
• Best for ultra-low-memory devices.

X-Ray_Alpha-iq3_m.gguf

• IQ3_M quantization, offering a medium block size for better accuracy.
• Suitable for low-memory devices.

X-Ray_Alpha-q4_0.gguf

• Pure Q4_0 quantization, optimized for ARM devices.
• Best for low-memory environments.
• Prefer IQ4_NL for better accuracy.

🚀 If you find these models useful

❤ Please click "Like" if you find this useful!
Help me test my AI-Powered Network Monitor Assistant with quantum-ready security checks:
👉 Free Network Monitor

💬 How to test:

1. Click the chat icon (bottom right on any page)
2. Choose an AI assistant type:
   • TurboLLM (GPT-4-mini)
   • FreeLLM (Open-source)
   • TestLLM (Experimental CPU-only)

What I’m Testing

I’m pushing the limits of small open-source models for AI network monitoring, specifically:

• Function calling against live network services
• How small can a model go while still handling:
  • Automated Nmap scans
  • Quantum-readiness checks
  • Metasploit integration

🟡 TestLLM – Current experimental model (llama.cpp on 6 CPU threads):

• ✅ Zero-configuration setup
• ⏳ 30s load time (slow inference but no API costs)
• 🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!

Other Assistants

🟢 TurboLLM – Uses gpt-4-mini for:

• Real-time network diagnostics
• Automated penetration testing (Nmap/Metasploit)
• 🔑 Get more tokens by downloading our Free Network Monitor Agent

🔵 HugLLM – Open-source models (≈8B params):

• 2x more tokens than TurboLLM
• AI-powered log analysis
• 🌐 Runs on Hugging Face Inference API

💡 Example AI Commands to Test:

1. "Give me info on my websites SSL certificate"
2. "Check if my server is using quantum safe encyption for communication"
3. "Run a quick Nmap vulnerability test"

X-Ray_Alpha

---

Click here for TL;DR Why it's important How can YOU help? How to RUN IT

---

This is a pre-alpha proof-of-concept of a real fully uncensored vision model.

Why do I say "real"? The few vision models we got (qwen, llama 3.2) were "censored," and their fine-tunes were made only to the text portion of the model, as training a vision model is a serious pain.

The only actually trained and uncensored vision model I am aware of is ToriiGate; the rest of the vision models are just the stock vision + a fine-tuned LLM.

Does this even work?

YES!

---

Why is this Important?

Having a fully compliant vision model is a critical step toward democratizing vision capabilities for various tasks, especially image tagging. This is a critical step in both making LORAs for image diffusion models, and for mass tagging images to pretrain a diffusion model.

In other words, having a fully compliant and accurate vision model will allow the open source community to easily train both loras and even pretrain image diffusion models.

Another important task can be content moderation and classification, in various use cases there might not be black and white, where some content that might be considered NSFW by corporations, is allowed, while other content is not, there's nuance. Today's vision models do not let the users decide, as they will straight up refuse to inference any content that Google \ Some other corporations decided is not to their liking, and therefore these stock models are useless in a lot of cases.

What if someone wants to classify art that includes nudity? Having a naked statue over 1,000 years old displayed in the middle of a city, in a museum, or at the city square is perfectly acceptable, however, a stock vision model will straight up refuse to inference something like that.

It's like in many "sensitive" topics that LLMs will straight up refuse to answer, while the content is publicly available on Wikipedia. This is an attitude of cynical patronism, I say cynical because corporations take private data to train their models, and it is "perfectly fine", yet- they serve as the arbitrators of morality and indirectly preach to us from a position of a suggested moral superiority. This gatekeeping hurts innovation badly, with vision models especially so, as the task of tagging cannot be done by a single person at scale, but a corporation can.

How can YOU help?

This is sort of "Pre-Alpha", a proof of concept, I did A LOT of shortcuts and "hacking" to make this work, and I would greatly appreciate some help to make it into an accurate and powerful open tool. I am not asking for money, but well-tagged data. I will take the burden and costs of the compute on myself, but I cannot do tagging at a large scale by myself.

Bottom line, I need a lot of well-tagged, diverse data

So:

• If you have well-tagged images
• If you have a link to a well-tagged image dataset
• If you can, and willing to do image tagging

Then please send an email with [DATASET] in the title to:

[email protected]

As you probably figured by the email address name, this is not my main email, and I expect it to be spammed with junk, so please use the [DATASET] tag so I can more easily find the emails of the good people who are actually trying to help.

Please see this dataset repo if you want to help:

X-Ray_Community_Tagging

Also, if you don't want to upload it to the repo (although it's encouraged, and you can protect it with a password for privacy), you can still help by linking a google drive or attach the images with the corrected output via the provided email.

Let's make this happen. We can do it!

---

TL;DR

• Fully uncensored and trained there's no moderation in the vision model, I actually trained it.
• The 2nd uncensored vision model in the world, ToriiGate being the first as far as I know, and this one is the second.
• In-depth descriptions very detailed, long descriptions.
• The text portion is somewhat uncensored as well, I didn't want to butcher and fry it too much, so it remain "smart".
• NOT perfect This is a POC that shows that the task can even be done, a lot more work is needed.
• Good Roleplay & Writing I used a massive corpus of high quality human (~60%) and synthetic data.

---

How to run it:

VRAM needed for FP16: 15.9 GB

Run inference with this

This is a pre-alpha POC (Proof Of Concept)

Instructions:

clone:

git clone https://github.com/SicariusSicariiStuff/X-Ray_Vision.git
cd X-Ray_Vision/

Settings up venv, (tested for python 3.11, probably works with 3.10)

python3.11 -m venv env
source env/bin/activate

Install dependencies

pip install git+https://github.com/huggingface/[email protected]
pip install torch
pip install pillow
pip install accelerate

Running inference

Usage:

python xRay-Vision.py /path/to/model/ /dir/with/images/

The output will print to the console, and export the results into a dir named after your image dir with the suffix "_TXT"

So if you run:

python xRay-Vision.py /some_path/x-Ray_model/ /home/images/weird_cats/

The results will be exported to:

/home/images/weird_cats_TXT/

---

Your support = more models

My Ko-fi page (Click here)

---

Citation Information

@llm{X-Ray_Alpha,
  author = {SicariusSicariiStuff},
  title = {X-Ray_Alpha},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/SicariusSicariiStuff/X-Ray_Alpha}
}

---

Other stuff

• X-Ray_Vision Easy stand-alone bulk vision inference at scale (inference a folder of images).
• SLOP_Detector Nuke GPTisms, with SLOP detector.
• LLAMA-3_8B_Unaligned The grand project that started it all.
• Blog and updates (Archived) Some updates, some rambles, sort of a mix between a diary and a blog.