Agent-almanac analyze-generative-diffusion-model

install
source · Clone the upstream repo
git clone https://github.com/pjt222/agent-almanac
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/pjt222/agent-almanac "$T" && mkdir -p ~/.claude/skills && cp -r "$T/i18n/wenyan-lite/skills/analyze-generative-diffusion-model" ~/.claude/skills/pjt222-agent-almanac-analyze-generative-diffusion-model-612c50 && rm -rf "$T"
manifest: i18n/wenyan-lite/skills/analyze-generative-diffusion-model/SKILL.md
source content

析生成式擴散模型

藉量化品質指標、噪聲排程檢查、交叉注意力圖分析與潛空間探查,評估預訓練之生成式擴散模型,以解模型行為、診斷失敗模式並指引微調決策。

適用時機

  • 以標準指標評預訓練之生成式擴散模型之輸出品質
  • 為生成圖像集計算 FID、IS、CLIP score 或 precision/recall
  • 經 SNR 曲線檢查並比對噪聲排程(線性、餘弦、學習)
  • 提取交叉注意力圖以解文圖之 token-區域對應
  • 於潛碼間插值,或於潛空間中發現語意方向
  • 為擴散模型管線偵測分佈外之輸入

輸入

  • 必要:預訓練模型識別碼或檢查點路徑(如
    stabilityai/stable-diffusion-2-1
  • 必要:分析模式——一或多:
    metrics
    schedule
    attention
    latent
  • 必要:指標計算之參考資料集(真實圖像或資料集名)
  • 選擇性:注意力分析之文字提詞(預設:模型適用之測試提詞)
  • 選擇性:指標計算之生成樣本數(預設:10000)
  • 選擇性:裝置配置(預設:若可用則
    cuda
    ,否則
    cpu

步驟

步驟一:量化評估

對參考資料集計算標準生成品質指標。

  1. 設置評估管線:
import torch
from diffusers import StableDiffusionPipeline
from torchmetrics.image.fid import FrechetInceptionDistance
from torchmetrics.image.inception import InceptionScore

device = "cuda" if torch.cuda.is_available() else "cpu"
pipe = StableDiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16
).to(device)

fid = FrechetInceptionDistance(feature=2048, normalize=True).to(device)
inception = InceptionScore(normalize=True).to(device)
  1. 將真實圖像饋入指標累積器:
from torch.utils.data import DataLoader

for batch in DataLoader(real_dataset, batch_size=64):
    imgs = (batch * 255).byte().to(device)
    fid.update(imgs, real=True)
  1. 生成樣本並累積偽統計量:
prompts = load_evaluation_prompts("prompts.txt")  # one prompt per line
n_generated = 0
while n_generated < 10000:
    prompt_batch = prompts[n_generated:n_generated + 8]
    images = pipe(prompt_batch, num_inference_steps=50).images
    tensors = torch.stack([to_tensor(img) for img in images]).to(device)
    byte_imgs = (tensors * 255).byte()
    fid.update(byte_imgs, real=False)
    inception.update(byte_imgs)
    n_generated += len(images)
  1. 計算文圖對齊之 CLIP score:
from torchmetrics.multimodal.clip_score import CLIPScore

clip_metric = CLIPScore(model_name_or_path="openai/clip-vit-large-patch14").to(device)
for prompt, image_tensor in zip(sampled_prompts, sampled_tensors):
    clip_metric.update(image_tensor.unsqueeze(0), [prompt])

print(f"FID: {fid.compute():.2f}")
print(f"IS:  {inception.compute()[0]:.2f} +/- {inception.compute()[1]:.2f}")
print(f"CLIP: {clip_metric.compute():.2f}")
  1. 為模式覆蓋計算 precision 與 recall:
from torchmetrics.image import FrechetInceptionDistance

# Precision: fraction of generated images near real manifold
# Recall: fraction of real images near generated manifold
# Use improved precision/recall (Kynkaanniemi et al., 2019) via
# feature embeddings from the Inception network

預期: 訓練良好之 Stable Diffusion 模型於標準基準上 FID 低於 30。ImageNet 類提詞之 IS 高於 50。文條件模型之 CLIP score 高於 25。Precision 與 recall 皆高於 0.6。

失敗時: 若 FID 高於 100,驗真實與生成圖像之解析度與正規化是否一致。若 CLIP score 低而 FID 可接受,模型生成貌似合理之圖像但不合提詞——查文字編碼器。為穩定 FID 估計確保至少 10,000 樣本。

步驟二:噪聲排程檢查

視覺化並比對前向與反向之噪聲排程。

  1. 自模型提取排程參數:
scheduler = pipe.scheduler
betas = torch.tensor(scheduler.betas) if hasattr(scheduler, 'betas') else None
alphas_cumprod = torch.tensor(scheduler.alphas_cumprod)
timesteps = torch.arange(len(alphas_cumprod))
  1. 計算信噪比曲線:
import numpy as np
import matplotlib.pyplot as plt

snr = alphas_cumprod / (1 - alphas_cumprod)
log_snr = torch.log(snr)

fig, axes = plt.subplots(1, 3, figsize=(18, 5))
axes[0].plot(timesteps.numpy(), alphas_cumprod.numpy())
axes[0].set_xlabel("Timestep"); axes[0].set_ylabel("alpha_cumprod")
axes[0].set_title("Cumulative Signal Retention")

axes[1].plot(timesteps.numpy(), log_snr.numpy())
axes[1].set_xlabel("Timestep"); axes[1].set_ylabel("log(SNR)")
axes[1].set_title("Log Signal-to-Noise Ratio")

if betas is not None:
    axes[2].plot(timesteps.numpy(), betas.numpy())
    axes[2].set_xlabel("Timestep"); axes[2].set_ylabel("beta")
    axes[2].set_title("Beta Schedule")
fig.tight_layout()
fig.savefig("noise_schedule.png", dpi=150)
  1. 比對多種排程類型:
from diffusers import DDPMScheduler

schedules = {
    "linear": DDPMScheduler(beta_schedule="linear", num_train_timesteps=1000),
    "cosine": DDPMScheduler(beta_schedule="squaredcos_cap_v2", num_train_timesteps=1000),
}

fig, ax = plt.subplots(figsize=(10, 6))
for name, sched in schedules.items():
    ac = torch.tensor(sched.alphas_cumprod)
    snr = torch.log(ac / (1 - ac))
    ax.plot(snr.numpy(), label=name)
ax.set_xlabel("Timestep"); ax.set_ylabel("log(SNR)")
ax.set_title("Schedule Comparison"); ax.legend()
fig.savefig("schedule_comparison.png", dpi=150)

預期: 餘弦排程於中段時步較線性呈更漸進之 SNR 降。log-SNR 曲線應自約 +10(潔)至 -10(純噪)。學習之排程應單調下降。

失敗時: 若 alphas_cumprod 非單調下降,排程設置有誤。若值為常,查解算器是否以模型之 config 正確初始化。對自訂解算器,驗

set_timesteps()
已被呼叫。

步驟三:注意力圖分析

自文條件模型提取並視覺化交叉注意力圖。

  1. 於 U-Net 之交叉注意力層註冊注意力 hook:
attention_maps = {}

def hook_fn(name):
    def fn(module, input, output):
        # Cross-attention: Q from image, K/V from text
        if hasattr(module, 'processor'):
            attention_maps[name] = output.detach().cpu()
    return fn

for name, module in pipe.unet.named_modules():
    if 'attn2' in name and hasattr(module, 'processor'):
        module.register_forward_hook(hook_fn(name))
  1. 行推論並於特定時步收集注意力:
prompt = "a red car parked next to a blue house"
timestep_attention = {}

# Custom callback to capture attention at specific timesteps
def callback_fn(pipe, step_index, timestep, callback_kwargs):
    if step_index in [5, 15, 30, 45]:
        timestep_attention[int(timestep)] = {
            k: v.clone() for k, v in attention_maps.items()
        }
    return callback_kwargs

output = pipe(prompt, num_inference_steps=50, callback_on_step_end=callback_fn)
  1. 視覺化 token-區域對應:
tokenizer = pipe.tokenizer
tokens = tokenizer.encode(prompt)
token_strings = [tokenizer.decode([t]) for t in tokens]

# Select a mid-resolution attention layer
layer_key = [k for k in attention_maps if 'mid' in k or 'up.1' in k][0]
attn = attention_maps[layer_key]  # shape: (batch, heads, hw, seq_len)
attn_avg = attn.mean(dim=1)  # average across heads
res = int(attn_avg.shape[1] ** 0.5)
attn_map = attn_avg[0].reshape(res, res, -1)

fig, axes = plt.subplots(2, min(len(token_strings), 6), figsize=(18, 6))
for idx, token in enumerate(token_strings[:6]):
    for row, (ts, ts_attn) in enumerate(list(timestep_attention.items())[:2]):
        a = ts_attn[layer_key].mean(dim=1)[0]
        a_res = int(a.shape[0] ** 0.5)
        axes[row, idx].imshow(a[:, idx].reshape(a_res, a_res), cmap="hot")
        axes[row, idx].set_title(f"t={ts}: '{token}'")
        axes[row, idx].axis("off")
fig.suptitle("Cross-Attention Maps by Token and Timestep")
fig.tight_layout()
fig.savefig("attention_maps.png", dpi=150)

預期: 內容 token("car"、"house")激活局部空間區域。風格/顏色 token("red"、"blue")激活與其關聯物件相疊之區。早時步(高噪)呈瀰散之注意力;晚時步呈銳而局部之注意力。

失敗時: 若一切注意力圖呈均勻,hook 或捕到自我注意力而非交叉注意力——驗層名含

attn2
(交叉)而非
attn1
(自我)。若注意力捕到然維度有誤,查輸出 tensor 索引是否合該層之 head 數與空間解析度。

步驟四:潛空間探查

藉插值與方向發現探潛空間之結構。

  1. 將參考圖像編碼至潛空間:
from diffusers import AutoencoderKL
from PIL import Image
import torchvision.transforms as T

vae = pipe.vae
transform = T.Compose([T.Resize(512), T.CenterCrop(512), T.ToTensor(),
                       T.Normalize([0.5], [0.5])])

def encode_image(image_path):
    img = transform(Image.open(image_path).convert("RGB")).unsqueeze(0).to(device)
    with torch.no_grad():
        latent = vae.encode(img.half()).latent_dist.sample() * vae.config.scaling_factor
    return latent

z1 = encode_image("image_a.png")
z2 = encode_image("image_b.png")
  1. 行球面線性插值(slerp):
def slerp(z1, z2, alpha):
    """Spherical linear interpolation between two latent codes."""
    z1_flat = z1.flatten()
    z2_flat = z2.flatten()
    omega = torch.acos(torch.clamp(
        torch.dot(z1_flat, z2_flat) / (z1_flat.norm() * z2_flat.norm()), -1, 1
    ))
    if omega.abs() < 1e-6:
        return (1 - alpha) * z1 + alpha * z2
    return (torch.sin((1 - alpha) * omega) * z1 + torch.sin(alpha * omega) * z2) / torch.sin(omega)

alphas = torch.linspace(0, 1, 8)
interpolated = [slerp(z1, z2, a.item()) for a in alphas]
decoded = []
for z in interpolated:
    with torch.no_grad():
        img = vae.decode(z / vae.config.scaling_factor).sample
    decoded.append(img.cpu())
  1. 經提詞對之差發現語意方向:
def get_text_embedding(prompt):
    tokens = pipe.tokenizer(prompt, return_tensors="pt", padding="max_length",
                            max_length=77, truncation=True).input_ids.to(device)
    with torch.no_grad():
        emb = pipe.text_encoder(tokens).last_hidden_state
    return emb

pos_emb = get_text_embedding("a happy person smiling")
neg_emb = get_text_embedding("a sad person frowning")
direction = pos_emb - neg_emb  # semantic direction in text embedding space
  1. 偵測分佈外之潛碼:
# Compute latent space statistics from a reference set
ref_latents = torch.stack([encode_image(p) for p in reference_paths])
ref_mean = ref_latents.mean(dim=0)
ref_std = ref_latents.std(dim=0)

def ood_score(z):
    """Mahalanobis-like OOD score (higher = more unusual)."""
    deviation = ((z - ref_mean) / (ref_std + 1e-6)).flatten()
    return deviation.norm().item()

test_z = encode_image("test_image.png")
score = ood_score(test_z)
print(f"OOD score: {score:.2f} (reference mean: {np.mean([ood_score(r) for r in ref_latents]):.2f})")

預期: 插值圖像呈順、語意有意義之轉換而無產物。語意方向加於多樣潛碼時產一致之屬性變化。分佈內圖像之 OOD 分數聚緊;異常者顯著更高。

失敗時: 若插值產糢糊或不連貫之中點,用 slerp 而非線性插值——線性插值橫越高維潛空間之低密度區域。若語意方向無可見效果,增方向幅度或驗文字編碼器與訓練時所用同。

驗證

  • FID 於至少 10,000 生成樣本與配對之真實樣本數計算
  • CLIP score 以訓練時所用同一 CLIP 模型計算(若適用)
  • 噪聲排程視覺化呈單調下降之 alphas_cumprod
  • Log-SNR 跨整時步範圍約自 +10 至 -10
  • 注意力圖於中解析度層解出每 token 之空間激活
  • 注意力自早(瀰散)至晚(局部)時步銳化
  • 潛插值順而無突跳或產物
  • OOD 偵測基線自至少 100 參考樣本立

常見陷阱

  • 解析度不符之 FID:真實與生成圖像於饋入 Inception 前須同解析度。否則 FID 失真
  • 遺正規化於 torchmetrics
    FrechetInceptionDistance(normalize=True)
    期 [0, 1] float tensor。若
    normalize=False
    則期 [0, 255] uint8。混慣例致 FID 無意義
  • hook 自我注意力而非交叉注意力:U-Net 中名
    attn1
    之層為自我注意力(圖對圖)。
    attn2
    為交叉注意力(文對圖)。混之產無資訊之均勻圖
  • 高維中之線性插值:二高維高斯間之線性插值經低密度殼。擴散模型之潛空間插值恆用 slerp
  • 忽 VAE scaling factor:Stable Diffusion 之潛碼於編碼後乘以
    vae.config.scaling_factor
    。遺施或除此因子致解碼圖像錯亂
  • precision/recall 樣本過少:少於每集 5,000 樣本之 precision 與 recall 估計不可靠。為穩定估計用至少 10,000

相關技能

  • implement-diffusion-network
    — 構建此技能評估之擴散模型
  • analyze-diffusion-dynamics
    — 此處檢查之噪聲過程之數學基礎
  • fit-drift-diffusion-model
    — 共享 SDE 基礎之另一擴散模型族