minor: debugging prints

app.py

@@ -4,7 +4,7 @@ import gradio as gr
 import torch
 import logging
 from diffusers import DiffusionPipeline
-from transformers import LlamaForCausalLM, PreTrainedTokenizerFast
+from transformers import LlamaForCausalLM, PreTrainedTokenizerFast, BitsAndBytesConfig
 from transformer_hidream_image import HiDreamImageTransformer2DModel
 from pipeline_hidream_image import HiDreamImagePipeline
 from schedulers.fm_solvers_unipc import FlowUniPCMultistepScheduler
@@ -12,6 +12,8 @@ from schedulers.flash_flow_match import FlashFlowMatchEulerDiscreteScheduler
 
 import subprocess
 
+print(f"Is CUDA available: {torch.cuda.is_available()}")
+print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
 try:
     print(subprocess.check_output(["nvcc", "--version"]).decode("utf-8"))
 except:
@@ -32,6 +34,7 @@ RESOLUTION_OPTIONS = [
     "1248 × 832 (Landscape)",
     "832 × 1248 (Portrait)"
 ]
+quantization_config = BitsAndBytesConfig(load_in_4bit=True)
 
 MODEL_PREFIX = "azaneko"
 LLAMA_MODEL_NAME = "hugging-quants/Meta-Llama-3.1-8B-Instruct-GPTQ-INT4"
@@ -69,6 +72,7 @@ pipe = HiDreamImagePipeline.from_pretrained(
     tokenizer_4=tokenizer_4,
     text_encoder_4=text_encoder_4,
     torch_dtype=torch.bfloat16,
+    quantization_config=quantization_config
 )
 pipe.transformer = transformer
 log_vram("✅ Pipeline loaded!")