bilingual hierarchical resource catalog indexing and navigation
Organizes 300+ LLM ecosystem resources across 25+ categories using a bilingual (Chinese/English) hierarchical markdown structure deployed via Jekyll GitHub Pages. The catalog uses a consistent section pattern with category headers, resource links, and descriptions that enable both human browsing and programmatic discovery through GitHub's raw markdown API. Each resource is tagged with domain (foundation, deployment, multimodal, etc.) enabling cross-domain navigation and filtering.
Unique: Uses a bilingual hierarchical organization (Chinese-first naming convention) across 25+ domain categories (Foundation & Training, RAG Systems, Agentic RL, Multimodal Systems, etc.) with 1,278-line single-file architecture enabling GitHub Pages deployment without backend infrastructure. Integrates DeepWiki architectural analysis to provide technical context for each category section.
vs alternatives: More comprehensive and domain-specific than Papers with Code or Hugging Face Model Hub for LLM ecosystem discovery; bilingual support and architectural depth analysis differentiates from English-only awesome lists.
foundation and training resource aggregation with data-to-model pipeline mapping
Catalogs 40+ resources spanning data processing, model training, fine-tuning frameworks, and reinforcement learning approaches. The catalog maps the complete pipeline from raw data curation through foundation model training, including tools for data annotation (Label Studio, Argilla), preprocessing (Hugging Face Datasets), fine-tuning (Unsloth, LLaMA-Factory), and agentic RL (veRL, AReaL). Resources are organized by training methodology (supervised fine-tuning, RLHF, DPO, GRPO) enabling builders to select appropriate frameworks for their training objectives.
Unique: Uniquely maps agentic reinforcement learning frameworks (veRL, AReaL, slime, Agent Lightning) alongside traditional fine-tuning, reflecting the shift toward reasoning model training. Includes specialized sections for GRPO (Group Relative Policy Optimization) and reasoning model training pipelines used in DeepSeek-R1 replication.
vs alternatives: More comprehensive than Papers with Code for training infrastructure; includes both data processing and RL training frameworks in one taxonomy, whereas most resources separate these concerns.
advanced reasoning and o1/o3 model resource aggregation
Catalogs 15+ resources for advanced reasoning models (OpenAI o1, o3, DeepSeek-R1) and open-source reasoning model implementations. The catalog maps how reasoning models differ from standard LLMs (chain-of-thought training, test-time compute, verification), including training approaches (GRPO, RL-based reasoning) and inference patterns. Resources span both commercial reasoning APIs and open-source implementations, enabling builders to understand and implement advanced reasoning capabilities.
Unique: Focuses specifically on advanced reasoning models (o1, o3, DeepSeek-R1) and their training approaches (GRPO, RL-based reasoning), reflecting the emerging frontier of reasoning-focused LLMs. Includes both commercial APIs and open-source implementations, enabling builders to understand and replicate reasoning capabilities.
vs alternatives: Uniquely focused on reasoning model training and implementation; most LLM resources treat reasoning as a capability of standard models rather than a distinct model category.
small and efficient model resource aggregation with optimization technique mapping
Catalogs 25+ small and efficient LLM models (Phi, TinyLlama, Mistral 7B, Qwen, Gemma) organized by optimization approach: quantization (GPTQ, AWQ, GGUF), distillation, pruning, and architectural efficiency. The catalog maps how efficient models trade off capability for size/speed, including benchmarks on standard tasks. Resources span both pre-optimized models and optimization frameworks, enabling builders to select or create efficient models for resource-constrained deployments.
Unique: Organizes efficient models by optimization approach (quantization, distillation, pruning, architectural efficiency) rather than just model name. Includes both pre-optimized models (Phi, TinyLlama) and optimization frameworks, reflecting the spectrum from ready-to-use to custom optimization.
vs alternatives: More optimization-technique-focused than individual model documentation; enables builders to understand efficiency tradeoffs and select or create efficient models matching their constraints.
model context protocol (mcp) resource aggregation with integration pattern guidance
Catalogs resources for Model Context Protocol (MCP), a standardized protocol for LLM context management and tool integration. The catalog maps MCP implementations, client libraries, and server implementations, including integration patterns with LLM applications. Resources span both MCP specification documentation and practical implementations, enabling builders to understand and implement MCP-based context management and tool orchestration.
Unique: Focuses specifically on Model Context Protocol (MCP) as a standardized approach to context management and tool integration, distinct from custom tool calling implementations. Maps MCP specification, client libraries, and server implementations, reflecting the emerging standardization of LLM context protocols.
vs alternatives: Uniquely focused on MCP standardization; most LLM resources treat tool integration as framework-specific rather than protocol-based.
learning resources aggregation spanning books, courses, and technical papers
Catalogs 50+ learning resources organized by format: books (LLM fundamentals, prompt engineering, RAG), courses (university courses, online platforms), and technical papers (foundational research, recent advances). The catalog maps resources by topic (transformer architecture, fine-tuning, agents, multimodal) and audience level (beginner, intermediate, advanced), enabling learners to find appropriate educational materials for their background and goals.
Unique: Organizes learning resources by format (books, courses, papers) and topic (transformers, fine-tuning, agents, multimodal) rather than just listing materials. Includes both foundational resources and cutting-edge research papers, reflecting the breadth of LLM knowledge.
vs alternatives: More topic-and-format-focused than general learning platforms; enables learners to find specific educational materials for their background and goals.
interactive demo and model arena discovery for comparative evaluation
Catalogs 10+ interactive platforms (Hugging Face Spaces, OpenRouter, Chatbot Arena, Together Playground) enabling side-by-side model comparison and evaluation. The catalog maps how platforms enable comparative evaluation (same prompt across models, user voting, leaderboards) and integration with multiple model providers. Resources span both community-driven arenas (Chatbot Arena) and commercial platforms (OpenRouter), enabling builders to evaluate models before integration.
Unique: Focuses on interactive platforms enabling side-by-side model comparison and community-driven evaluation, distinct from automated benchmarking. Includes both community arenas (Chatbot Arena) and commercial platforms (OpenRouter), reflecting the spectrum from open to managed evaluation.
vs alternatives: More interactive-and-comparative-focused than static benchmarks; enables real-time model evaluation and community-driven quality assessment.
inference and serving framework discovery with deployment pattern guidance
Aggregates 30+ inference serving frameworks (vLLM, TensorRT-LLM, SGLang, Ollama, LM Studio) organized by deployment pattern (local, cloud, edge, batch). The catalog maps frameworks to specific optimization techniques (quantization, batching, KV-cache optimization) and hardware targets (CPU, GPU, mobile). Resources include both open-source inference engines and commercial serving platforms, enabling builders to select frameworks matching their latency, throughput, and cost requirements.
Unique: Organizes inference frameworks by deployment pattern (local, cloud, edge, batch) rather than just framework name, with explicit mapping to optimization techniques (quantization, batching, KV-cache) and hardware targets. Includes both open-source engines (vLLM, SGLang, Ollama) and commercial platforms (Together AI, Replicate).
vs alternatives: More deployment-pattern-focused than framework-specific documentation; enables builders to find solutions by use case (low-latency API, batch processing, edge deployment) rather than learning individual framework APIs.
+7 more capabilities