Skip to content
View NidAll's full-sized avatar

Highlights

  • Pro

Block or report NidAll

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
NidAll/README.md

Nidhal // NidAll

computer engineer · AI systems · Linux · inference optimization

Linux Python CUDA

I like making heavyweight AI models behave on hardware they were never supposed to fit on.


⚡ What I build

Project What it does Focus
ComfyUI-Qwen3-TTS-Quant Native ComfyUI runtime and streaming quantization toolkit for the full Qwen3-TTS 12 Hz 1.7B family W8A8 ConvRot · packed INT8 · CUDA graphs · static KV caches
comfyui-mixed-quantizer Diffusion and text-encoder quantization with stable W4A8 and experimental mixed-precision planning W4A8 · W4A4 · INT8 · validation · memory-aware planning
OpenCode 2 support for Ask OpenCode 2 compatibility work for the Ask terminal AI client Rust · CLI integration · permissions · read-only agent design

I spend a lot of time around the point where model architecture, kernels, quantization, memory pressure, and runtime behavior collide.


🧠 Engineering interests

AI systems: quantization, low-VRAM inference, CUDA kernels, memory residency, KV-cache design, graph capture, attention backends, and runtime optimization.

Inference: local LLMs, diffusion models, speech generation, multimodal systems, custom quantized runtimes, and squeezing maximum performance from constrained hardware.

Linux: rolling-release systems, terminal tooling, performance engineering, containers, and understanding the stack all the way down.

Software engineering: Rust, Python, systems integration, reproducible tooling, testing, and contributing fixes upstream.


🛠️ Current environment

OS          CachyOS
Desktop     KDE Plasma
Terminal    Kitty
Shell       Fish

AI          ComfyUI · llama.cpp · PyTorch
GPU Stack   CUDA · FlashAttention · Comfy Kitchen
Languages   Python · Rust · C/C++
Dev         Git · GitHub · Docker · OpenCode
Terminal    micro · mdfried · ltop · lazydocker

I like tools that are fast, composable, transparent, and close to the system.


🔬 What I enjoy optimizing

model
  ↓
architecture
  ↓
quantization
  ↓
kernel
  ↓
memory layout
  ↓
runtime
  ↓
benchmark
  ↓
repeat

If a model technically shouldn't fit, that usually makes it more interesting.


🐧 Beyond code

Linux is the playground. AI systems are the obsession.

I also enjoy philosophy, mathematics, anime & manga, storytelling, and two objectively excellent animals:

   _~_        /\_/\
  (o o)      ( o.o )
 /  V  \      > ^ <
/(  _  )\
  ^^ ^^

 penguin       cat

My completely scientific rating scale:

Ω > Z > S > A > B > C > D > F


📊 GitHub

NidAll GitHub streak

Build it. Measure it. Understand why it works.

Linux · AI · CUDA · Quantization · Systems

Pinned Loading

  1. ComfyUI-Qwen3-TTS-Quant ComfyUI-Qwen3-TTS-Quant Public

    Production-focused ComfyUI nodes and a streaming quantization toolkit for fast, memory-efficient Qwen3-TTS VoiceDesign inference with W8A8 ConvRot, CUDA kernels, static KV caching, and runtime diag…

    Python

  2. ComfyUI_PromptStyler ComfyUI_PromptStyler Public

    Custom ComfyUI node that applies curated style prefix/suffix templates to your positive prompt and outputs CONDITIONING for KSampler.

    Python 18 2

  3. TranslateGemma-Studio TranslateGemma-Studio Public

    Local TranslateGemma UI with a thin llama.cpp adapter, streaming translations, long-text chunking, and a clean light/dark interface all running offline on your PC.

    JavaScript 16 1

  4. LuxTTS-Gradio LuxTTS-Gradio Public

    Gradio UI and beginner setup for LuxTTS voice cloning, with full advanced controls and fast local inference.

    Python 38 4

  5. LLaDa-Image_ComfyUI LLaDa-Image_ComfyUI Public

    Forked from RealRebelAI/LLaDa-Image_ComfyUI

    a custom node set to run LLaDa-Image in ComfyUI

    Python

  6. comfyui-native-quantizer comfyui-native-quantizer Public

    Native mixed-precision quantization for ComfyUI models and text encoders W4A8, W4A4, INT8, FP8, streaming conversion, architecture-aware policies, and validation.

    Python