全部能力

Persistent cross-session memory using Memento MCP knowledge graph (mcp__memento__* tools). Recall-before-acting: search memory before starting tasks, on errors, and when receiving corrections. Multi-dimensional search: two queries per recall event (technical topic + process/workflow learnings). Store-after-discovery: persist solutions, conventions, and corrections immediately. Three-step recall: search, open_nodes, traverse relations. WORKING_STATE.md for crash recovery. Self-reminder protocol every 5-10 messages. Activate on task start, errors, corrections, session boundaries, or explicit memory requests. See references/agents-md-setup.md for AGENTS.md integration.

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2×…

Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6×…

Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training

Control LLM output with regex and grammars, guarantee valid JSON/XML/code generation, enforce structured formats, and build multi-step workflows with Guidance…

当用户明确要求"写/改研究内容""研究内容+创新+年度计划编排"时使用。为 NSFC 正文"（二）研究内容"写作/重构，并同步编排"特色与创新"和"三年年度研究计划"，输出可直接落到 LaTeX 模板的三个 extraTex 文件。

GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible…

Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting…

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited…

README 优先的 AI 仓库复现主编排器。当用户需要端到端、最小可信的复现流程，且该流程会读取仓库…

Extract structured data from LLM responses with Pydantic validation, retry failed extractions automatically, parse complex JSON with type safety, and stream…

Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximize…

Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL from…

创建、审阅、更新和搜索 ADK 文档的指南 - 当用户询问编写、维护或审计 ADK 机器人文档时使用

High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from…

Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use…

Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+…

Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or…

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense…

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward…