Instructions to use Akiyue/awwl with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Akiyue/awwl with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Akiyue/awwl", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
| import json | |
| import matplotlib.pyplot as plt | |
| import os | |
| import numpy as np | |
| # Danh sách các folder bạn muốn so sánh | |
| # (Tên folder phải khớp với cái bạn đã chạy) | |
| experiments = [ | |
| {"path": "ablat_best", "label": "AWWL (Best: a0.8, p2.0)", "color": "red"}, | |
| {"path": "ablat_static", "label": "Static Loss (p0.0)", "color": "gray"}, | |
| {"path": "ablat_alpha05", "label": "Balanced (a0.5)", "color": "blue"}, | |
| {"path": "ablat_power1", "label": "Linear Decay (p1.0)", "color": "orange"}, | |
| ] | |
| # Hàm làm mượt (Smoothing) vì loss nhảy lung tung khó nhìn | |
| def smooth(scalars, weight=0.95): | |
| last = scalars[0] | |
| smoothed = list() | |
| for point in scalars: | |
| smoothed_val = last * weight + (1 - weight) * point | |
| smoothed.append(smoothed_val) | |
| last = smoothed_val | |
| return smoothed | |
| plt.figure(figsize=(10, 6)) | |
| for exp in experiments: | |
| json_path = os.path.join(exp["path"], "loss_history.json") | |
| if os.path.exists(json_path): | |
| with open(json_path, "r") as f: | |
| data = json.load(f) | |
| losses = data["losses"] | |
| # Chỉ vẽ nếu có dữ liệu | |
| if len(losses) > 0: | |
| # Làm mượt loss để biểu đồ đẹp như trong báo | |
| smoothed_losses = smooth(losses, weight=0.99) # 0.99 là rất mượt | |
| # Tạo trục X (Steps) | |
| steps = range(len(smoothed_losses)) | |
| plt.plot(steps, smoothed_losses, label=exp["label"], color=exp["color"], linewidth=1.5) | |
| print(f"Loaded {exp['label']}: {len(losses)} steps") | |
| else: | |
| print(f"Warning: File not found {json_path}") | |
| plt.title("Training Loss Convergence Comparison") | |
| plt.xlabel("Training Steps") | |
| plt.ylabel("Loss (Smoothed)") | |
| plt.legend() | |
| plt.grid(True, alpha=0.3) | |
| plt.yscale("log") # Dùng scale log nếu chênh lệch lớn, hoặc bỏ dòng này nếu muốn linear | |
| plt.tight_layout() | |
| # Lưu biểu đồ ra ảnh | |
| plt.savefig("loss_comparison_chart.png", dpi=300) | |
| print("✅ Chart saved to loss_comparison_chart.png") | |
| plt.show() |