""" MOSS-Image — EntroPack pre-quantized inference example. Downloads the pre-quantized packages and their json sidecars from the `DiffSynth-Studio/MOSS-Image-EntroPack` model repo (each json is a ready-to-use `MODEL_CONFIGS` entry), then runs inference. Defaults to SFT INT8 6bit; switch `dit_package` / `text_encoder_package` (and `base_model_id` for Base) to use another level. Available packages (paths inside the prequant model repo): dit: models/MOSS-Image-SFT/dit_3.5bit.safetensors mixed allocation, 3.5bit models/MOSS-Image-SFT/dit_4.5bit.safetensors mixed allocation, 4.5bit models/MOSS-Image-SFT/dit_6bit.safetensors 6bit models/MOSS-Image-SFT/dit_int8_6bit.safetensors hybrid INT8 + 8.5bit sensitive, 6bit models/MOSS-Image-SFT/dit_int8_7bit.safetensors hybrid INT8 + 8.5bit sensitive, 7bit models/MOSS-Image-Base/dit_4.5bit.safetensors mixed allocation, 4.5bit models/MOSS-Image-Base/dit_6bit.safetensors 6bit models/MOSS-Image-Base/dit_int8_6bit.safetensors hybrid INT8 + 8.5bit sensitive, 6bit models/MOSS-Image-Base/dit_int8_7bit.safetensors hybrid INT8 + 8.5bit sensitive, 7bit text_encoder: models/text_encoder_4.5bit.safetensors 4.5bit models/text_encoder_6bit.safetensors 6bit models/text_encoder_int8_6bit.safetensors hybrid INT8 + 8.5bit sensitive, 6bit models/text_encoder_int8_7bit.safetensors hybrid INT8 + 8.5bit sensitive, 7bit The text encoder packages are shared between Base and SFT. `dit_package` and `text_encoder_package` are chosen independently: any combination works (e.g. `dit_int8_6bit` with `text_encoder_4.5bit`). """ import json import torch from diffsynth.configs import MODEL_CONFIGS from diffsynth.pipelines.moss_image import MOSSImagePipeline, ModelConfig prequant_model_id = "DiffSynth-Studio/MOSS-Image-EntroPack" base_model_id = "openmoss/MOSS-Image-SFT" dit_package = "models/MOSS-Image-SFT/dit_int8_7bit.safetensors" text_encoder_package = "models/text_encoder_int8_7bit.safetensors" def register_prequantized(package_pattern): """Download the pre-quantized package and its json sidecar (a ready-to-use MODEL_CONFIGS entry), return a loadable ModelConfig.""" package = ModelConfig(model_id=prequant_model_id, origin_file_pattern=package_pattern) package.download_if_necessary() sidecar = ModelConfig(model_id=prequant_model_id, origin_file_pattern=package_pattern.replace(".safetensors", ".json")) sidecar.download_if_necessary() with open(sidecar.path) as handle: entry = json.load(handle) if not any(config.get("model_hash") == entry["model_hash"] for config in MODEL_CONFIGS): MODEL_CONFIGS.append(entry) return package pipe = MOSSImagePipeline.from_pretrained( torch_dtype=torch.bfloat16, device="cuda", model_configs=[ register_prequantized(dit_package), register_prequantized(text_encoder_package), ModelConfig(model_id=base_model_id, origin_file_pattern="vae/diffusion_pytorch_model.safetensors", computation_dtype=torch.float32), ], tokenizer_config=ModelConfig(model_id=base_model_id, origin_file_pattern="tokenizer/"), ) # Optional: torch.compile acceleration. pipe.compile_pipeline() prompt = '''A wide view of a refrigerator beside a kitchen counter with a fruit bowl. A small sticky note on the refrigerator door is handwritten in Chinese: "记得买牛奶和鸡蛋! 牛奶两盒,鸡蛋一打。 顺便带一袋面包。"''' image = pipe(prompt, height=1024, width=1024, seed=42, num_inference_steps=50, cfg_scale=3.0) image.save("image_MOSS-Image-EntroPack.jpg")