flux-lora-nga379

FLUX.2 Klein 9B LoRA training artifacts for trigger word nga379.

Base model: black-forest-labs/FLUX.2-klein-base-9B

Training config:

• LoRA rank/alpha: 32/32
• Batch size: 4
• Save/sample interval: 300 steps
• Current latest checkpoint: 4500
• Resume target in config: 6000 steps
• Low VRAM: false
• Quantization: false

Files included:

• config/nga379_flux2_klein9b_lora.yml: AI Toolkit config updated to 6000 steps.
• output/nga379_flux2_klein9b_lora/config.yaml: output config snapshot updated to 6000 steps.
• output/nga379_flux2_klein9b_lora/nga379_flux2_klein9b_lora_000001800.safetensors through nga379_flux2_klein9b_lora_000004500.safetensors.
• output/nga379_flux2_klein9b_lora/optimizer.pt: optimizer state from the latest saved 4500-step checkpoint.
• output/nga379_flux2_klein9b_lora/samples/: sample images from step 1800 onward.

To resume in AI Toolkit, restore these files into the same relative paths under /root/ai-toolkit, make sure the dataset is available at /root/dataset/nga379, then run:

cd /root/ai-toolkit
source venv/bin/activate
python run.py config/nga379_flux2_klein9b_lora.yml