Added Kvikontent's Midjourney V6 Stable Diffusion LoRA Model

[benliang99]

This update integrates Kvikontent's Midjourney V6 Stable Diffusion LoRA model into the BitMind Subnet's collection of synthetic image generators.

Model Configuration

The following configuration has been added to config.py to support the new model:

bitmind-subnet/bitmind/validator/config.py

Lines 181 to 193 in baf7a87

	"runwayml/stable-diffusion-v1-5-midjourney-v6": {
	"pipeline_cls": StableDiffusionPipeline,
	"from_pretrained_args": {
	"model_id": "runwayml/stable-diffusion-v1-5",
	"use_safetensors": True,
	"torch_dtype": torch.float16,
	},
	"lora_model_id": "Kvikontent/midjourney-v6",
	"lora_loading_args": {
	"use_peft_backend": True
	},
	"enable_model_cpu_offload": False
	},

LoRA Model Weight Loading

Support for loading LoRA model weights using the Parameter-Efficient Fine-Tuning (PEFT) library has also been added to the synthetic data generator:

bitmind-subnet/bitmind/synthetic_data_generation/synthetic_data_generator.py

Lines 422 to 429 in baf7a87

	# Load LoRA weights if specified
	if 'lora_model_id' in model_config:
	bt.logging.info(f"Loading LoRA weights from {model_config['lora_model_id']}")
	lora_loading_args = model_config.get('lora_loading_args', {})
	self.model.load_lora_weights(
	model_config['lora_model_id'],
	**lora_loading_args
	)

[benliang99]

Consolidated in #171