LukeG89

Results 83 comments of LukeG89

Import failed Searge LLM for ComyUI v1.0 - Custom Node ID 93

@aprisma2008 You are missing `llama_cpp_python`, a dependency that Searge LLM needs to run the models. The link provided by Searge node can install it for Python 3.11, but the desktop...

Import failed Searge LLM for ComyUI v1.0 - Custom Node ID 93

Have you managed on installing llama-cpp-python?

AMD RX 580 help?

You should try this ComfyUI version, I see there are installation instructions for old cards: https://github.com/patientx/ComfyUI-Zluda

[Performance Regression] GGUF models run noticeably slower with only ~50% VRAM usage after removing WAN Block Swap node (compared to previous manual offloading behavior at 80-90% VRAM)

It's not just with GGUF models (and not just Wan). I just made some tests, I was about to open a new bug report, but I guess I could post...

[Performance Regression] GGUF models run noticeably slower with only ~50% VRAM usage after removing WAN Block Swap node (compared to previous manual offloading behavior at 80-90% VRAM)

> Do you have a comparison basis for an older version or is it just about the lora performance discrepancy? No, unfortunately I don't have a comparison with previous versions....

[Performance Regression] GGUF models run noticeably slower with only ~50% VRAM usage after removing WAN Block Swap node (compared to previous manual offloading behavior at 80-90% VRAM)

> @LukeG89 these options are performance critical, especially pinned memory. If you have a problem with these options left on we can take that as a straight bug report Never...

Given groups=1, weight of size [4, 4, 1, 1], expected input[1, 16, 64, 64] to have 4 channels, but got 16 channels instead

One of the models is probably wrong, I guess it's the VAE (suggestion: you need to better rename the models)

Given groups=1, weight of size [4, 4, 1, 1], expected input[1, 16, 64, 64] to have 4 channels, but got 16 channels instead

Try using this VAE: https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/ae.safetensors Here if you need clip L and T5xxl text encoders: https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

Given groups=1, weight of size [4, 4, 1, 1], expected input[1, 16, 64, 64] to have 4 channels, but got 16 channels instead

Anyway, LoRA is also not compatible, this is why you have all those `WARNING - lora key not loaded`

live preview is not working in the ksampler node

Now they moved the option. ### On portable/manual installation #### Go to Settings and search "preview" _____ ### On Desktop app #### Go to Settings > Server-Config and find Preview