fix headings, H2->H1

Author: ev-br

docs/source/ecosystem_overview/architectural.md

@@ -1,4 +1,4 @@
-## The Architectural Imperative: Performance Beyond Frameworks
+# The Architectural Imperative: Performance Beyond Frameworks
 
 As model architectures converge—for example, on multimodal Mixture-of-Experts (MoE) Transformers—the pursuit of peak performance is leading to the emergence of "Megakernels." A Megakernel is effectively the entire forward pass (or a large portion) of one specific model, hand-coded using a lower-level API like the CUDA SDK on NVIDIA GPUs. This approach achieves maximum hardware utilization by aggressively overlapping compute, memory, and communication. Recent work from the research community has demonstrated that this approach can yield significant throughput gains, over 22% in some cases, for inference on GPUs. This trend is not limited to inference; evidence suggests that some large-scale training efforts have involved low-level hardware control to achieve substantial efficiency gains.
 

docs/source/ecosystem_overview/comparative.md

@@ -1,4 +1,4 @@
-## A Comparative Perspective: The JAX/TPU Stack as a Compelling Choice
+# A Comparative Perspective: The JAX/TPU Stack as a Compelling Choice
 
 The modern Machine Learning landscape offers many excellent, mature toolchains. The JAX AI Stack, however, presents a unique and compelling set of advantages for developers focused on large-scale, high-performance ML, stemming directly from its modular design and deep hardware co-design.
 

docs/source/ecosystem_overview/conclusion.md

@@ -1,4 +1,4 @@
-## Conclusion: A Durable, Production-Ready Platform for the Future of AI 
+# Conclusion: A Durable, Production-Ready Platform for the Future of AI 
 
 The data provided in the table above draws to a rather simple conclusion \- these stacks have their own strengths and weaknesses in a small number of areas but overall are vastly similar from the software standpoint. Both stacks provide out of the box turnkey solutions for pre-training, post-training adaptation and deployment of foundational models.
 

docs/source/ecosystem_overview/core.md

@@ -1,4 +1,4 @@
-## The Core JAX AI Stack
+# The Core JAX AI Stack
 
 The core JAX AI Stack consists of five key libraries that provide the foundation for model development: JAX, [Flax](https://flax.readthedocs.io/en/stable/), [Optax](https://optax.readthedocs.io/en/latest/), [Orbax](https://orbax.readthedocs.io/en/latest/) and [Grain](https://google-grain.readthedocs.io/en/latest/).
 

docs/source/ecosystem_overview/extended.md

@@ -1,4 +1,4 @@
-## The Extended JAX Ecosystem
+# The Extended JAX Ecosystem
 
 Beyond the core stack, a rich ecosystem of specialized libraries provides the infrastructure, advanced tools, and application-layer solutions needed for end-to-end ML development.
 

docs/source/ecosystem_overview/modular.md

@@ -1,4 +1,4 @@
-## A Modular, Compiler-First Architecture for Modern AI
+# A Modular, Compiler-First Architecture for Modern AI
 
 The [JAX AI stack](https://jaxstack.ai/) extends the JAX numerical core with a collection of Google-backed composable libraries, evolving it into a robust, end-to-end, open-source platform for Machine Learning at extreme scales. As such, the JAX AI stack consists of a comprehensive and robust ecosystem that addresses the entire ML lifecycle: