Open-source ComfyUI nodes losslessly compress models: 28GB to 19GB, bit-identical
New-Shift6661 · reddit · 2026-09-24
A developer released ComfyUI-LosslessCompression, ComfyUI nodes that compress AI model files losslessly — weights stay bit-for-bit identical in any dtype (fp32/fp16/bf16/fp8), so inference results match the original exactly.
The key insight: in trained weights the sign and mantissa bits are near-random; only exponent bits are predictable. The tool stores sign/mantissa unchanged and gives each tensor's exponents short codes, landing within 1–2 points of the theoretical limit — e.g. a 28GB bf16 model shrinks to about 19GB. Usage is simple: open a new ComfyUI tab, pick a model, hit RUN; runtime scales with model size.
Repo: GitHub - GDI-Forces/ComfyUI-LosslessCompression
More from Infra
- A Ready-to-Use Prompt That Makes Your Agent Audit Its Own API Bills — gethackteam · 2026-09-24
- ~50us per kernel launch possible, but only by forking a custom single-model inference stack — AlpinDale · 2026-09-24
- Tencent Hunyuan extends critical-batch-size theory to LLM RL: 29% faster GRPO, 2.29× PPO throughput — TencentHunyuan · 2026-09-24
- NVIDIA demos Nemotron 3.5 Lightning running locally on DGX Spark and Station — NVIDIA Developer · 2026-09-24
- For LLM workloads, AMD CCD count matters more than core count: 16-core tops out at 125GB/s — HankYeomans · 2026-09-24
- If 1 billion people ran personal AI agents, CPU and memory demands would be staggering — firstadopter · 2026-09-24