Cedric Weber

[Cedric Weber @weber_cedric]

@cocktailpeanut @gantasmo How is the quality compared to Ace Step 1.5?

[Cedric Weber @weber_cedric]

@Teknium after every "recreate/re-pull image" in my docker/portainer set up for Hermes I get:
Dropping root privilegeschown: changing ownership of '/opt/data/config.yaml': Operation not permitted

i have to"sudo chown -R 1000:1000 ..." to get is working again:
Any idea for a fix?

[Cedric Weber @weber_cedric]

@Zyra_exe @Teknium @tonbistudio Yeah had the struggle in the beginning too.
U need to set the ID for the free_response_channel and maybe set require_mention: false

[Cedric Weber @weber_cedric]

@Teknium Hermes only works again after I restore the permissions in the console with "chmod 666 /opt/data/config.yaml" and restart the container. (3/3)

[Cedric Weber @weber_cedric]

@Teknium I think I've noticed an error with the new dashboard.
I'm running Hermes via a Docker Container (Portainer) on my NAS. Today I wanted to test the dashboard and, for example, change the TTS model in my config via the dashboard and save it. (1/2)

[Cedric Weber @weber_cedric]

@Teknium But when I update the config via the dashboard, it resets the permissions and the logs show Warning: Failed to load config: [Errno 13] Permission denied: '/opt/data/config.yaml'"
Along with that, "⚠️ Provider authentication failed" in Telegram and Discord. (2/3)

[Cedric Weber @weber_cedric]

@SOL_mfer @Teknium @rileybrown I read somewhere on Reddit
Llama ccp- Gemma 31B Q8 with f16 KV

[Cedric Weber @weber_cedric]

@NetworkChuck @NousResearch I am so looking forward to an in-depth video from you about Hermes!

[Cedric Weber @weber_cedric]

@umb_c @toyxyz3 I noticed with the nvfp4 you rather need 14 steps than 8 steps when using the distilled lora
Or need to to set the strength of the lora to 0.9 or even 1.0

[Cedric Weber @weber_cedric]

@cocktailpeanut Yes I am, would love a fix for this for my MBP M2

[Cedric Weber @weber_cedric]

@AIWarper I just tried it and think it‘s better compared to VibeVoice

Especially for cloning german voices

[Cedric Weber @weber_cedric]

@AitestAi10100 @ostrisai @Alibaba_Qwen Which quantization settings did you end up using in the AI Toolkit? And did u gather some experience yet?

[Cedric Weber @weber_cedric]

@AIWarper Top for movement and bottom for lighting

[Cedric Weber @weber_cedric]

@NVIDIA_AI_PC Open Source Models

[Cedric Weber @weber_cedric]

@ostrisai Going for Wan2.2 because it has video but is also a great image model. So with Wan we can do both.

[Cedric Weber @weber_cedric]

@fofrAI I bet this Lora serves as a really good base for an Image to 3D model, is the best open source model still Hunyuan or did I miss something?

[Cedric Weber @weber_cedric]

@fofrAI The hands are so bad ☹️

[Cedric Weber @weber_cedric]

@WhateverrrrVip @cocktailpeanut There is a standalone version with which Flux works. It is probably still due to the fact that pytorch on Windows has not yet received an update for the new architecture.
But when it's running, the 5090 is insanely fast. From my MacBook M2, a huge leap in speed tho.

[Cedric Weber @weber_cedric]

@hahaitsmike @cocktailpeanut @andyorsow @finegrain_ai @philz1337x I guess u have 🙃