Environment:
OS:Linux Mint 22.3 - Cinnamon 64-bit
CPU:AMD Ryzen 5 8400F
GPU: RTX 5060 8G VRAM
RAM: 6000MHz 16GB RAM
version installed: diffsynth-2.1.7
minimax h3 Model files:
~/DiffSynth-Studio/models/DiffSynth-Studio/MiniMax-H3-NF4$ ls
audio_vae_nf4.safetensors
minimax-h3-ref2va-pruned-nf4.safetensors
minimax-h3-text-encoder-nf4.safetensors
minimax_h3_turbo_v4_step600_ema.safetensors
video_vae_nf4.safetensors
minimax h3 lora link https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora
Symptom:
i want to run minimax-h3-ref2va-pruned-nf4 but with turbo lora minimax_h3_turbo_v4_step600_ema with num_inference_steps=6 . Note default num_inference_steps=50 so for 5 second video it would take avg 40 minutes for my hardware so that's why i want to try turbo lora decrease num of steps to speed up generation speed and decrease time
so i asked gemini to add that and he did and that is script it made "minimax_h3.py" and it failed see logs for more info.
so i want to know can i use lora with pruned version and how to run lora with minimax-h3-ref2va-pruned-nf4 model
notice the diffrence is 7 gb between the pruned and non pruned versions
original low vram script MiniMax-H3-NF4-Pruned-Ref2VA.py
also how can i provide and add referenced photo and audio to use in model inside python script
Also i noticed in https://diffsynth-studio-doc.readthedocs.io/en/latest/Model_Details/MiniMax-H3.html there is mention of using lightx2v turbo version but not for ref2va so can i use this turbo lora maybe edit its name inside this script minimax_h3/model_inference_low_vram/MiniMax-H3-FL2VA-Turbo.py
https://www.modelscope.cn/models/lightx2v/Minimax-h3-Turbo/file/view/master/minimax_h3_ref2v_turbo_4step_v0.1_bf16.safetensors
https://github.com/modelscope/DiffSynth-Studio/blob/main/examples/minimax_h3/model_inference_low_vram/MiniMax-H3-FL2VA-Turbo.py
MiniMax-H3-NF4-Pruned-Ref2VA.py
log.txt
minimax_h3.py
Environment:
OS:Linux Mint 22.3 - Cinnamon 64-bit
CPU:AMD Ryzen 5 8400F
GPU: RTX 5060 8G VRAM
RAM: 6000MHz 16GB RAM
version installed: diffsynth-2.1.7
minimax h3 Model files:
~/DiffSynth-Studio/models/DiffSynth-Studio/MiniMax-H3-NF4$ ls
audio_vae_nf4.safetensors
minimax-h3-ref2va-pruned-nf4.safetensors
minimax-h3-text-encoder-nf4.safetensors
minimax_h3_turbo_v4_step600_ema.safetensors
video_vae_nf4.safetensors
minimax h3 lora link https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora
Symptom:
i want to run minimax-h3-ref2va-pruned-nf4 but with turbo lora minimax_h3_turbo_v4_step600_ema with num_inference_steps=6 . Note default num_inference_steps=50 so for 5 second video it would take avg 40 minutes for my hardware so that's why i want to try turbo lora decrease num of steps to speed up generation speed and decrease time
so i asked gemini to add that and he did and that is script it made "minimax_h3.py" and it failed see logs for more info.
so i want to know can i use lora with pruned version and how to run lora with minimax-h3-ref2va-pruned-nf4 model
notice the diffrence is 7 gb between the pruned and non pruned versions
original low vram script MiniMax-H3-NF4-Pruned-Ref2VA.py
also how can i provide and add referenced photo and audio to use in model inside python script
Also i noticed in https://diffsynth-studio-doc.readthedocs.io/en/latest/Model_Details/MiniMax-H3.html there is mention of using lightx2v turbo version but not for ref2va so can i use this turbo lora maybe edit its name inside this script minimax_h3/model_inference_low_vram/MiniMax-H3-FL2VA-Turbo.py
https://www.modelscope.cn/models/lightx2v/Minimax-h3-Turbo/file/view/master/minimax_h3_ref2v_turbo_4step_v0.1_bf16.safetensors
https://github.com/modelscope/DiffSynth-Studio/blob/main/examples/minimax_h3/model_inference_low_vram/MiniMax-H3-FL2VA-Turbo.py
MiniMax-H3-NF4-Pruned-Ref2VA.py
log.txt
minimax_h3.py