Update README.md

README.md

@@ -78,7 +78,7 @@ convert_to_quant -i $1 \
   --custom-type mxfp8 \
   --custom-layers "layers\.(10|16|26)\.attention\.qkv\.weight|layers\.(27|28)\.attention\.qkv\.weight|layers\.(0|1)\.attention\.out\.weight|layers\.(3|6|9|11|12|13|14|19|20|26)\.attention\.out\.weight|layers\.(27|28|29)\.attention\.out\.weight|layers\.(3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26)\.feed_forward\.w1\.weight|layers\.(27|28|29)\.feed_forward\.w1\.weight|layers\.(16|17|18|19|20|21)\.adaLN_modulation\.0\.weight|context_refiner\.(0|1)\.attention\.qkv\.weight|context_refiner\.(0|1)\.feed_forward\.w1\.weight|context_refiner\.(0|1)\.feed_forward\.w2\.weight|context_refiner\.(0|1)\.feed_forward\.w3\.weight|noise_refiner\.(0)\.attention\.(qkv|out)\.weight|noise_refiner\.(0)\.feed_forward\.(w1|w2)\.weight|noise_refiner\.(1)\.feed_forward\.w1\.weight" \
   --exclude-layers "layers\.(29)\.attention\.qkv\.weight|layers\.(22|23|24|25|26)\.adaLN_modulation\.0\.weight|layers\.(27|28|29)\.adaLN_modulation\.0\.weight|context_refiner\.(0|1)\.attention\.out\.weight|context_refiner\.(1)\.feed_forward\.w2\.weight|noise_refiner\.(1)\.attention\.qkv\.weight|noise_refiner\.(1)\.attention\.out\.weight|noise_refiner\.(1)\.feed_forward\.w2\.weight|noise_refiner\.(0|1)\.feed_forward\.w3\.weight" \
-  --num-iter 6000 --top-p 0.35 --calib-samples 8192 --manual-seed 42 \
+  --num-iter 6000 --top-p 0.35 --calib-samples 8192 \
   --scale-optimization iterative --scale-refinement-rounds 2 \
   --extract-lora --lora-rank 32 \
   -o "${1%%.safetensors}-nvfp4.safetensors"
@@ -97,7 +97,7 @@ Use the LoRA at **1.5–2.0** strength in ComfyUI for maximum fidelity.
 
 ## Requirements
 
-- **Inference**: CUDA 13.0+, PyTorch 2.8+, [`comfy-kitchen`](https://github.com/Comfy-Org/comfy-kitchen), Blackwell GPU (RTX 50xx / B100 / B200)
+- **Inference**: CUDA 13.0+, PyTorch 2.10+, [`comfy-kitchen`](https://github.com/Comfy-Org/comfy-kitchen), Blackwell GPU (RTX 50xx / B100 / B200)
 - **Generation**: `convert_to_quant >= 1.2.6`, `comfy-kitchen`
 
 ---
@@ -137,7 +137,7 @@ Recommendations were cross-referenced against the DiT quantization literature:
 
 ## Credits
 
-- Layer sensitivity analysis via [`quant_probe`](https://github.com/insecure-erasure/quant_probe)
 - Quantization engine: [`convert_to_quant`](https://github.com/silveroxides/convert_to_quant) by silveroxides
 - Z-Image Turbo model by [Tongyi-MAI](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo)
-- ComfyUI integration via [`comfy-kitchen`](https://github.com/Comfy-Org/comfy-kitchen)
+- ComfyUI integration via [`comfy-kitchen`](https://github.com/Comfy-Org/comfy-kitchen)
+- Layer sensitivity analysis via [`quant_probe`](https://github.com/insecure-erasure/quant_probe)