Skip to content

Commit 8f29083

Browse files
authored
sync(zh): 上游交互图表系统全量同步(133 课 figure + 13 渲染模块) (#24)
上游 PR #279(b963cf63..148de36)curriculum-wide interactive figure system: - A 类:133 个课程 docs/zh.md 各插入 1 个 ```figure 块,锚点与上游 en.md 一致(108 个在「动手构建」前、23 个在「实际使用」前、 2 个在「架构」前);figure 名属代码不翻。 debugging-neural-networks 类型 Practice→Build 随上游。 - B 类:13 个新 figures-*.js 渲染模块 + lesson-figures.js 升级 (LF 工具集 svgEl/select/clamp/lerp/raf + 新渲染器)1:1 照搬上游。 - C 类:lesson.html 仅手动加 13 个 script 引用 + 版本号 20260610e, zh 特化(findH2/Analytics/SEO)未动。 - 跳过上游 data.js 构建产物,本仓 node site/build.js 重建。 - .sync-upstream-base 推进至 148de36。 验证:figure 名序列 en/zh 全 133 文件一致;137 个 figure 名全部有 渲染器;本地冒烟页 137/137 挂载渲染成功;build --check 503 课通过。
1 parent 31de49b commit 8f29083

150 files changed

Lines changed: 7048 additions & 7 deletions

File tree

  • phases
    • 01-math-foundations
      • 01-linear-algebra-intuition/docs
      • 02-vectors-matrices-operations/docs
      • 03-matrix-transformations/docs
      • 04-calculus-for-ml/docs
      • 05-chain-rule-and-autodiff/docs
      • 06-probability-and-distributions/docs
      • 07-bayes-theorem/docs
      • 08-optimization/docs
      • 09-information-theory/docs
      • 10-dimensionality-reduction/docs
      • 11-singular-value-decomposition/docs
      • 12-tensor-operations/docs
      • 13-numerical-stability/docs
      • 14-norms-and-distances/docs
      • 16-sampling-methods/docs
      • 17-linear-systems/docs
      • 18-convex-optimization/docs
      • 19-complex-numbers/docs
      • 20-fourier-transform/docs
      • 21-graph-theory/docs
      • 22-stochastic-processes/docs
    • 02-ml-fundamentals
      • 02-linear-regression/docs
      • 03-logistic-regression/docs
      • 04-decision-trees/docs
      • 05-support-vector-machines/docs
      • 06-knn-and-distances/docs
      • 07-unsupervised-learning/docs
      • 08-feature-engineering/docs
      • 09-model-evaluation/docs
      • 10-bias-variance/docs
      • 12-hyperparameter-tuning/docs
      • 14-naive-bayes/docs
      • 17-imbalanced-data/docs
    • 03-deep-learning-core
      • 01-the-perceptron/docs
      • 02-multi-layer-networks/docs
      • 03-backpropagation/docs
      • 04-activation-functions/docs
      • 05-loss-functions/docs
      • 06-optimizers/docs
      • 07-regularization/docs
      • 08-weight-initialization/docs
      • 09-learning-rate-schedules/docs
      • 10-mini-framework/docs
      • 11-intro-to-pytorch/docs
      • 12-intro-to-jax/docs
      • 13-debugging-neural-networks/docs
    • 04-computer-vision
      • 01-image-fundamentals/docs
      • 02-convolutions-from-scratch/docs
      • 03-cnns-lenet-to-resnet/docs
      • 04-image-classification/docs
      • 15-real-time-edge/docs
    • 05-nlp-foundations-to-advanced
      • 01-text-processing/docs
      • 02-bag-of-words-tfidf/docs
      • 03-word-embeddings-word2vec/docs
      • 05-sentiment-analysis/docs
      • 06-named-entity-recognition/docs
      • 08-cnns-rnns-for-text/docs
      • 09-sequence-to-sequence/docs
      • 10-attention-mechanism/docs
      • 11-machine-translation/docs
      • 16-text-generation-pre-transformer/docs
      • 19-subword-tokenization/docs
    • 06-speech-and-audio
      • 01-audio-fundamentals/docs
      • 02-spectrograms-mel-features/docs
      • 11-real-time-audio-processing/docs
    • 07-transformers-deep-dive
      • 02-self-attention-from-scratch/docs
      • 03-multi-head-attention/docs
      • 04-positional-encoding/docs
      • 06-bert-masked-language-modeling/docs
      • 07-gpt-causal-language-modeling/docs
      • 12-kv-cache-flash-attention/docs
      • 13-scaling-laws/docs
      • 15-attention-variants/docs
    • 08-generative-ai
      • 02-autoencoders-vae/docs
      • 03-gans-generator-discriminator/docs
      • 06-diffusion-ddpm-from-scratch/docs
      • 07-latent-diffusion-stable-diffusion/docs
    • 09-reinforcement-learning
      • 01-mdps-states-actions-rewards/docs
      • 02-dynamic-programming/docs
      • 03-monte-carlo-methods/docs
      • 04-q-learning-sarsa/docs
      • 06-policy-gradients-reinforce/docs
    • 10-llms-from-scratch
      • 01-tokenizers/docs
      • 02-building-a-tokenizer/docs
      • 04-pre-training-mini-gpt/docs
      • 05-scaling-distributed/docs
      • 07-rlhf/docs
      • 08-dpo/docs
      • 10-evaluation/docs
      • 11-quantization/docs
      • 12-inference-optimization/docs
      • 13-building-complete-llm-pipeline/docs
      • 14-open-models-architecture-walkthroughs/docs
      • 16-differential-attention-v2/docs
      • 17-native-sparse-attention/docs
      • 19-dualpipe-parallelism/docs
      • 20-deepseek-v3-walkthrough/docs
      • 21-jamba-hybrid-ssm-transformer/docs
      • 22-async-hogwild-inference/docs
      • 25-speculative-decoding/docs
    • 11-llm-engineering
    • 12-multimodal-ai
      • 01-vision-transformer-patch-tokens/docs
      • 02-clip-contrastive-pretraining/docs
    • 13-tools-and-protocols/06-mcp-fundamentals/docs
    • 14-agent-engineering
      • 01-the-agent-loop/docs
      • 03-reflexion-verbal-rl/docs
      • 06-tool-use-and-function-calling/docs
      • 07-memory-virtual-context-memgpt/docs
    • 15-autonomous-systems
      • 01-long-horizon-agents/docs
      • 02-star-family-reasoning/docs
      • 07-recursive-self-improvement/docs
      • 10-claude-code-permission-modes/docs
      • 12-durable-execution/docs
    • 16-multi-agent-and-swarms
      • 01-why-multi-agent/docs
      • 05-supervisor-orchestrator-pattern/docs
    • 17-infrastructure-and-production
      • 02-inference-platform-economics/docs
      • 03-gpu-autoscaling-kubernetes/docs
      • 04-vllm-serving-internals/docs
      • 06-sglang-radixattention/docs
      • 07-tensorrt-llm-blackwell/docs
      • 08-inference-metrics-goodput/docs
      • 09-production-quantization/docs
      • 18-vllm-production-stack-lmcache/docs
      • 28-self-hosted-serving-selection/docs
    • 18-ethics-safety-alignment
      • 02-reward-hacking-goodhart/docs
      • 03-direct-preference-optimization-family/docs
    • 19-capstone-projects
      • 08-production-rag-chatbot/docs
      • 27-eval-harness-fixture-tasks/docs
      • 28-observability-otel-traces/docs
      • 49-lm-eval-harness/docs
  • site

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

.sync-upstream-base

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1 +1 @@
1-
b963cf63ffbb574af35da8db301ebb1381515ed8
1+
148de3663f8bab4c90355292de8d0fac81dc2a86

phases/01-math-foundations/01-linear-algebra-intuition/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -178,6 +178,10 @@ QR 分解内部就是这么干的。Q 是那组标准正交基,R 记录投影
178178
- 计算特征值(QR 算法)
179179
- 最小二乘回归(标准的数值方法)
180180

181+
```figure
182+
eigen-directions
183+
```
184+
181185
## 动手构建
182186

183187
### 第 1 步:从零写向量(Python)

phases/01-math-foundations/02-vectors-matrices-operations/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -113,6 +113,10 @@ Broadcasting stretches the vector across rows:
113113

114114
每个现代框架都会自动做这件事。理解它能让你在形状看起来不对、代码却照跑不误时不犯迷糊。
115115

116+
```figure
117+
vector-projection
118+
```
119+
116120
## 动手构建
117121

118122
### 第 1 步:Vector 类

phases/01-math-foundations/03-matrix-transformations/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -232,6 +232,10 @@ det = -1: area preserved but orientation flipped (reflection)
232232
| det(Reflection) | = -1 (orientation flipped)
233233
```
234234

235+
```figure
236+
matrix-transform
237+
```
238+
235239
## 动手构建
236240

237241
### 第 1 步:从零写变换矩阵(Python)

phases/01-math-foundations/04-calculus-for-ml/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -388,6 +388,10 @@ graph RL
388388

389389
前向传播算出预测和损失。反向传播算出损失对每个权重的梯度。然后每个权重往下坡迈一小步。重复几百万步。这就是深度学习。
390390

391+
```figure
392+
derivative-tangent
393+
```
394+
391395
## 动手构建
392396

393397
### 第 1 步:从零写数值导数

phases/01-math-foundations/05-chain-rule-and-autodiff/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -163,6 +163,10 @@ PyTorch 内部:
163163

164164
这个图是动态的(define-by-run)。每次前向传播都会构建一个新图。这就是为什么 PyTorch 支持在模型里写控制流(if/else、循环)。
165165

166+
```figure
167+
chain-rule
168+
```
169+
166170
## 动手构建
167171

168172
### 第 1 步:Value 类

phases/01-math-foundations/06-probability-and-distributions/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -254,6 +254,10 @@ Log-softmax 把 softmax 和 log 合在一起以保证数值稳定。PyTorch 在
254254

255255
从任意分布采样需要逆变换采样、拒绝采样或重参数化技巧(VAE 里用)这类技术。
256256

257+
```figure
258+
gaussian-pdf
259+
```
260+
257261
## 动手构建
258262

259263
### 第 1 步:概率基础

phases/01-math-foundations/07-bayes-theorem/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -195,6 +195,10 @@ MAP 在参数本身之上加了一个先验。如果你相信参数应该偏小
195195

196196
**模型比较是贝叶斯的。** 贝叶斯信息准则(BIC)、边际似然和贝叶斯因子,全都用贝叶斯推理在模型之间选择而不过拟合。
197197

198+
```figure
199+
bayes-update
200+
```
201+
198202
## 动手构建
199203

200204
### 第 1 步:贝叶斯定理函数

phases/01-math-foundations/08-optimization/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -195,6 +195,10 @@ graph TD
195195

196196
尖锐的最小值泛化得差。平坦的最小值泛化得好。这是带动量的 SGD 在最终测试准确率上常常胜过 Adam 的原因之一:它的噪声防止落进尖锐的最小值。
197197

198+
```figure
199+
gradient-descent
200+
```
201+
198202
## 动手构建
199203

200204
### 第 1 步:定义一个测试函数

phases/01-math-foundations/09-information-theory/docs/zh.md

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -273,6 +273,10 @@ Perplexity = e^H(P,Q) (if using nats)
273273

274274
GPT-2 在常见基准上达到约 30 的困惑度。现代模型在表示充分的领域里能做到个位数。
275275

276+
```figure
277+
entropy-kl
278+
```
279+
276280
## 动手构建
277281

278282
### 第 1 步:信息量和熵

0 commit comments

Comments
 (0)