MakazhanAlpamys / Soup
Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.
MakazhanAlpamys / Soup
Jit-Roy / WeeLLM
WeeLLM runs large diffusion models with as little as 4 GB of VRAM, without any quantization. It dynamically determines how many layers can fit within the available VRAM and streams the text encoder and transformer layers to the GPU layer by layer, enabling inference on hardware with limited VRAM. It supports both safetensors and GGUF models.
Thefrizzy1 / The_frizzy1-Workflows
Verified, organised docs for every The_frizzy1 low-VRAM ComfyUI workflow - GGUF image/video/audio from 4GB VRAM.
AcademiaSD / AcademiaSD_LoRAlab-Krea2
Low-VRAM LoRA Training for Krea-2
yalun753 / moe-l2
MoE expert offload for low-VRAM GPUs — run 100B+ MoE models (DeepSeek, Qwen, Mixtral) on 8 GB cards. Expert cache with LRU hot experts, OpenAI-compatible proxy, GGUF multi-shard.
AcademiaSD / AcademiaSD_LoRAlab-TrainerStudio
Train LoRAs for Qwen-Image 2.1, FLUX.2 Klein 9B, FLUX.3 Image Klein, Krea 2, Z-Image, MOSS-Image, ERNIE-Image, Ideogram 4, Anima, SDXL (Pony, Illustrious, NoobAI), LTX 2.3, MiniMax-H3, and more on NVIDIA GPUs from 4–8 GB VRAM. One install, one launcher, NF4 models. Windows, Linux and RunPod.
sepiablue-ai / ComfyUI-H3-X2-Stream
ComfyUI nodes for MiniMax H3: INT8 X2 VAE and asynchronous NVENC saving. 5-second 1088x1920 video with audio in about 60 s on an RTX 4070 12GB, about 30% faster than the FP16 X2 VAE.
kadenball / qwen38-27b-rtx3060-dcfr
RouteWeaver: experimental cache-safe speculative inference and CPU/GPU transfer optimization for Qwen 27B on low-VRAM NVIDIA GPUs.
1314521gjy / cyber-on-strata
在 8 GB 显存的显卡上跑 35B MoE（Cyber-Tiel-Coder-35B-A3B）：专家权重流式上卡、未命中专家由 CPU 池就地计算。Strata 引擎的 fork（MIT）。RTX 4080S 实测 8 GB 档 57.7–63.5 tok/s、全常驻 134–136 tok/s。Build from THIS repo — upstream does not support this model.