[docs] fix: update readme & image path

README.md

@@ -45,7 +45,7 @@
 
 > 🐉 LoongForge is part of Baidu Baige's **Loong** open-source series — named after the traditional Chinese **loong boat (龙舟)**, a symbol of coordinated power and forward momentum.
 
-**LoongForge** is a unified training framework for **LLMs, VLMs, VLAs, and diffusion models**, covering **pre-training**, **continued pre-training**, and **SFT**. Built upon Megatron-LM with deep systemic enhancements across **model coverage**, **training performance**, and **hardware support**, it delivers **significant speedups over mainstream open-source baselines**.
+**LoongForge** is a unified training framework for **LLMs, VLMs, diffusion, and embodied models**, covering **pre-training**, **continued pre-training**, and **SFT**. Built upon Megatron-LM with deep systemic enhancements across **model coverage**, **training performance**, and **hardware support**, it delivers **significant speedups over mainstream open-source baselines**.
 
 Before going open-source, LoongForge was developed as **AIAK-Training-LLM**, Baidu Baige's training acceleration stack. It has supported production training for enterprise customers across **Education**, **Computer Vision**, and **Embodied AI**, typically delivering **30%~50% speedup over customer baselines**, with the largest production runs reaching **5,000+ XPUs**.
 
@@ -95,7 +95,7 @@ See the full documentation for installation, tutorials, and advanced usage — [
 
 Measured on **v0.1.1** across LLM, VLM, VLA and DIT workloads against mainstream open-source training baselines:
 
-<img width="2422" height="1264" alt="image" src="https://github.com/user-attachments/assets/1553b7eb-d379-4e63-b293-2a716c4996be" />
+<img alt="LoongForge Benchmark Speedup" src="docs/assets/images/benchmark_speedup.png" />
 
 <details>
 <summary><b>📋 Detailed configurations & footnotes</b></summary>
@@ -109,7 +109,6 @@ Measured on **v0.1.1** across LLM, VLM, VLA and DIT workloads against mainstream
 | Qwen3-VL-30B-A3B | VLM | VeOmni<sup>†</sup> | 32 × A800<sup>‡</sup> · GBS 128 · 32K | **1.45×** |
 | GR00T N1.6 | VLA | LeRobot<sup>†</sup> | 8 × A800<sup>‡</sup> · GBS 128 · 224×224 | **2.31×** |
 | Pi0.5 | VLA | OpenPI<sup>†</sup> | 8 × A800<sup>‡</sup> · GBS 112 · 224×224 | **1.65×** |
-| Wan2.2 | DIT | DiffSynth<sup>†</sup> | 8 × A800<sup>‡</sup> · 480×832x49 | **2.16×** |
 
 > <sup>§</sup> Due to test-bed scale limits, **DeepSeek-V3.2** was validated separately on a reduced-layer configuration — LoongForge's **DSA CUDA kernel optimizations** still deliver **~5× speedup** over Megatron-LM and reach **64K sequence** (baseline OOMs beyond 8K).<br>
 > <sup>†</sup> Numbers reflect baseline and LoongForge versions at the time of measurement, and may evolve as implementations change.<br>

README_zh.md

@@ -45,7 +45,7 @@
 
 > 🐉 LoongForge 是百度百舸 **Loong** 开源系列的一员 —— 名字源于中国传统 **龙舟**，象征协同发力与破浪前行。
 
-**LoongForge** 是面向 **LLM、VLM、VLA 与 Diffusion 模型** 的统一训练框架，覆盖 **预训练（Pre-training）**、**持续预训练（Continued Pre-training）** 和 **SFT**。基于 Megatron-LM 在 **模型覆盖度**、**训练性能** 和 **硬件支持** 三个维度做了深度系统性增强，相对主流开源训练方案有**显著的性能提升**。
+**LoongForge** 是面向 **LLM、VLM、Diffusion 与 Embodied 模型** 的统一训练框架，覆盖 **预训练（Pre-training）**、**持续预训练（Continued Pre-training）** 和 **SFT**。基于 Megatron-LM 在 **模型覆盖度**、**训练性能** 和 **硬件支持** 三个维度做了深度系统性增强，相对主流开源训练方案有**显著的性能提升**。
 
 在开源之前，LoongForge 的前身是 **AIAK-Training-LLM** —— 百度百舸的训练加速栈，已在 **教育**、**计算机视觉** 和 **Embodied AI** 等多家企业客户的生产训练中落地，相对客户原有方案通常带来 **30%~50% 加速**，最大规模的生产训练任务达到 **5,000+ XPU**。
 
@@ -94,7 +94,7 @@
 ## 📊 性能 Benchmark
 
 在 **v0.1.1** 版本上针对 LLM、VLM、VLA、DIT 四类工作负载，与主流开源训练方案的对比结果：
-<img width="2422" height="1264" alt="image" src="https://github.com/user-attachments/assets/1553b7eb-d379-4e63-b293-2a716c4996be" />
+<img alt="LoongForge Benchmark Speedup" src="docs/assets/images/benchmark_speedup.png" />
 
 <details>
 <summary><b>📋 细节描述</b></summary>
@@ -108,7 +108,6 @@
 | Qwen3-VL-30B-A3B | VLM | VeOmni<sup>†</sup> | 32 × A800<sup>‡</sup> · GBS 128 · 32K | **1.45×** |
 | GR00T N1.6 | VLA | LeRobot<sup>†</sup> | 8 × A800<sup>‡</sup> · GBS 128 · 224×224 | **2.31×** |
 | Pi0.5 | VLA | OpenPI<sup>†</sup> | 8 × A800<sup>‡</sup> · GBS 112 · 224×224 | **1.65×** |
-| Wan2.2 | DIT | DiffSynth<sup>†</sup> | 8 × A800<sup>‡</sup> · 480×832x49 | **2.16×** |
 
 > <sup>§</sup> 受测试台规模限制，**DeepSeek-V3.2** 在减层配置下单独验证 —— LoongForge 的 **DSA CUDA Kernel 优化** 相对 Megatron-LM 仍带来 **~5× 加速**，并可支持 **64K 序列长度**（基线在 8K 以上即 OOM）。<br>
 > <sup>†</sup> 数据反映测量时对应基线的实现，后续可能随实现演进而变化。<br>

docs/assets/images/benchmark_image.py

@@ -24,7 +24,6 @@
 ROWS = [
     ("DeepSeek-V3.2 Lite", "MoE + DSA", "Megatron-LM", "Reduced layers · GBS 128 · 8K",      5.04, "§"),
     ("GR00T N1.6",         "VLA",       "LeRobot",     "8 × A800 · GBS 128 · 224×224",       2.31, ""),
-    ("Wan 2.2",            "DIT",       "DiffSynth",   "8 × A800 · 480×832×49",              2.16, ""),
     ("Pi 0.5",             "VLA",       "OpenPI",      "8 × A800 · GBS 112 · 224×224",       1.65, ""),
     ("Qwen3-VL-30B-A3B",   "VLM",       "VeOmni",      "32 × A800 · GBS 128 · 32K",          1.45, ""),
     ("Qwen3-30B-A3B",      "MoE",       "Megatron-LM", "32 × A800 · GBS 1024 · 32K",         1.16, ""),