Learning Paths
Skill paths are narrower than courses. Each one gathers the lessons and review checkpoints needed to build one concrete ML systems capability.
Focused Collection
This path focuses the catalog around one working skill.
A focused path for sampling controls, speculative decoding, Medusa, lookahead decoding, and DeepSeek MTP.
Open skill pathThis path focuses the catalog around one working skill.
A focused transformer path for implementing attention scores, self-attention, masks, and multi-head shape transforms.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for LoRA, quantization, large-model quantization diagnostics, quantization schemes, and MoE tradeoffs.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for understanding KV cache mechanics, shared-KV attention variants, efficient attention, and KV cache serving optimization.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for absolute, rotary, relative, ALiBi/T5-style, and long-context position methods.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for building a small end-to-end transformer training loop from tokens, embeddings, batches, layers, and decoding.
Open skill pathFocused Collection
This path focuses the catalog around one working skill.
A focused path for understanding the minimal architecture needed to start a coding agent, route events, cross service boundaries, and execute runtime actions.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for tracing conversation events, controller state transitions, and memory condensation boundaries.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for adding runtime capabilities and delegation behavior through OpenHands runtime components, memory, and microagents.
Open skill pathThis path focuses the catalog around one working skill.
A focused path for understanding how CodeAct and function calling convert model output into executable agent actions.
Open skill path