GLM-5.3-Flash-Uncensored

LoRA-merged exports of GLM-5.3-Flash. The adapter was merged with scale lora_alpha / r = 1.

Variants

• FP8/: the source checkpoint's 128x128 E4M3 block format. Quantized weights were dequantized, updated with the LoRA delta, and requantized per block.
• NVFP4/: compressed-tensors NVFP4A16, converted from the merged FP8 checkpoint through an FP8 block dequantizer.

Each directory contains a completion marker and a provenance manifest.

This repository inherits the MIT license of the base model.