schema = "m64/capsule@1"
description = "Kromoss + DMD2 8-step LoRA text-to-image on GB10 (diffusers). lodestones/Kromoss (MOSS-Image fine-tune) with its 8-step DMD2 LoRA: 1024x1024, 8 steps, CFG 1, shift 5, bf16. About 19 s per image on a DGX Spark (GB10)."
kinds = ["image.generate"]
authors = ["dh7"]
license = "apache-2.0"

[hardware]
platform = "linux-aarch64"
gpu = "NVIDIA GB10"
gpu_count = 1
# ~10 GB transformer + ~17 GB Qwen3.5-9B text encoder (bf16) + fp32 VAE and LoRA.
vram_gb = 40
driver = ">=580"
disk_gb = 40

[runtime]
sandbox = "container"
# No image: the Dockerfile builds on nvcr.io/nvidia/pytorch:25.09-py3 pinned by digest.
gpu = true
shm_gb = 8
internal_port = 9219
health = "/health"
# Measured: ~335 s to load off NFS in the source recipe; local NVMe is faster.
expected_load_seconds = 420
model_name = "kromoss-dmd2"
user = "65534:65534"

# Same file set as the original download script (which excludes diffusion_models/*,
# text_encoders/*, assets/* and *.webp: the ComfyUI repackings and the card images).
[[artifacts]]
type = "hf"
repo = "lodestones/Kromoss"
revision = "7f9bb7b06d7b6c27d294b0bab5a2b3056c379279"
include = [
  ".gitattributes", "LICENSE", "README.md", "model_index.json", "*.py",
  "8-steps-*.safetensors",
  "scheduler/*", "text_encoder/*", "tokenizer/*", "transformer/*", "vae/*",
]
mount = "/weights/kromoss"

[params]
size = "1024x1024"
num_inference_steps = 8
guidance_scale = 1.0
shift = 5.0
lora = "8-steps-dmd2-lora"


[routes]
extra = ["/v1/config"]