跳到正文- Voice Agents
- Transformer architecture
- Tokenization
- Test-Time Compute
- Structured Outputs
- Streaming
- Source: YaRN: Efficient Context Window Extension (Peng et al., 2023)
- Source: XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in LLMs
- Source: XGrammar
- Source: Why Language Models Hallucinate (Kalai et al., 2025)
- Source: Voice AI & Voice Agents: An Illustrated Primer
- Source: vLLM Expert Parallel Deployment docs
- Source: Vertex AI Model Garden — Explore Models
- Source: Unsloth Faster MoE training docs
- Source: Understand and count tokens (Gemini API docs)
- Source: Turning Up the Heat: Min-p Sampling (Nguyen et al., 2025)
- Source: Tulu 3 (Lambert et al., 2024) — RLVR
- Source: Towards Understanding Sycophancy in Language Models (Anthropic)
- Source: Tokenization counts: the impact of tokenization on arithmetic in frontier LLMs
- Source: Token counting (Claude API docs)
- Source: tiktoken (OpenAI)
- Source: The Open Source AI Definition 1.0 (OSI)
- Source: The Llama 4 herd (Meta AI blog)
- Source: The Curious Case of Neural Text Degeneration (Holtzman et al., 2020)
- Source: The "think" tool (Anthropic Engineering, 2025)
- Source: Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
- Source: SentencePiece (Kudo & Richardson, 2018)
- Source: Scaling LLM Test-Time Compute Optimally (Snell et al., 2024)
- Source: s1 — Simple Test-Time Scaling (Muennighoff et al., 2025)
- Source: RULER: What's the Real Context Size of Your Long-Context Language Models? (NVIDIA)
- Source: RouteLLM (Ong et al. 2024)
- Source: RoFormer: Rotary Position Embedding (Su et al., 2021)
- Source: Robust Speech Recognition via Large-Scale Weak Supervision (Whisper)
- Source: Reduce hallucinations (Claude Docs)
- Source: Reasoning models guide (OpenAI API docs)
- Source: Qwen3.8 model cards (Qwen, Hugging Face)
- Source: Qwen3 Technical Report and launch blog
- Source: Pricing (OpenAI API)
- Source: Pricing (Anthropic Claude API)
- Source: Playwright MCP README
- Source: Performant local mixture-of-experts CPU inference with GPU acceleration in llama.cpp
- Source: Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
- Source: OSWorld Benchmark
- Source: Optimizing for cost and intelligence (Anthropic)
- Source: OpenAI, Computer-Using Agent (CUA)
- Source: OpenAI Voice Agents Guide
- Source: OpenAI Text-to-Speech Guide
- Source: OpenAI Structured Outputs Guide
- Source: OpenAI Streaming API Responses Guide
- Source: OpenAI Speech-to-Text Guide
- Source: OpenAI Realtime Voice Activity Detection Guide
- Source: OpenAI Realtime API Guide
- Source: OpenAI Prompt Caching Guide
- Source: OpenAI Image Generation Guide
- Source: OpenAI Function Calling Guide
- Source: OpenAI deprecations page (search and deep research models)
- Source: OpenAI Computer Use Guide
- Source: OpenAI Batch API Guide
- Source: Notebook: How LLMs & transformers work
- Source: Non-determinism in GPT-4 is caused by Sparse MoE (152334H, 2023)
- Source: NoLiMa: Long-Context Evaluation Beyond Literal Matching (ICML 2025)
- Source: Neural Machine Translation of Rare Words with Subword Units (Sennrich et al., 2016)
- Source: Muse Spark Is Now an Open-Weights Model (PyTorch blog)
- Source: Models overview (Mistral docs)
- Source: Models and pricing (xAI docs)
- Source: Models and pricing (DeepSeek API docs)
- Source: Models (OpenAI API docs)
- Source: Mixture of Experts Explained (Hugging Face blog)
- Source: Mixtral of Experts
- Source: Mistral Large 4 model card (Mistral AI, Hugging Face)
- Source: Microsoft Foundry — Foundry Models Overview
- Source: Messages API reference (Claude API)
- Source: Lost in the Middle: How Language Models Use Long Contexts (Liu et al.)
- Source: Long context prompting for Claude 2.1 (Anthropic)
- Source: Llama Prompt Guard 2 model card
- Source: Llama Guard 4 model card
- Source: Llama 4 Community License and model page (Meta)
- Source: LiveKit, Using a Transformer to Improve End of Turn Detection
- Source: Let Me Speak Freely? (Format Restrictions Study)
- Source: Learning to Reason with LLMs (OpenAI, 2024)
- Source: Language Model Tokenizers Introduce Unfairness Between Languages (Petrov et al., 2023)
- Source: Kimi K3.1 Pro model card (Moonshot AI, Hugging Face)
- Source: Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena (Zheng et al.)
- Source: Inverse Scaling in Test-Time Compute (Gema et al., 2025)
- Source: Introducing Meta Llama 3 (Meta AI blog)
- Source: Hugging Face Transformers: Tokenization algorithms
- Source: How to make your completions outputs consistent with the new seed parameter (OpenAI Cookbook)
- Source: gpt-oss-120b model card (OpenAI, Hugging Face)
- Source: gpt-oss-120b & gpt-oss-20b Model Card (arXiv)
- Source: generateContent / GenerationConfig (Gemini API reference)
- Source: Gemma 4 model overview and card (Google)
- Source: Gemini models (Google AI Studio docs)
- Source: Gemini Live API Docs
- Source: Gemini Image Understanding Docs
- Source: Gemini Batch API Docs
- Source: Gemini API Structured Outputs Docs
- Source: Gemini API Streaming Interactions Docs
- Source: Gemini API pricing (Google)
- Source: Gemini API Function Calling Docs
- Source: Gemini API Docs — Thinking
- Source: Gemini API Context Caching and Pricing Docs
- Source: FrugalGPT (Chen, Zaharia, Zou 2023)
- Source: FlashAttention (Dao et al., 2022)
- Source: Efficient Guided Generation for Large Language Models
- Source: Effective context engineering for AI agents (Anthropic)
- Source: Do NOT Think That Much for 2+3=? (Chen et al., 2024)
- Source: Detecting hallucinations in large language models using semantic entropy (Nature, 2024)
- Source: Defeating Nondeterminism in LLM Inference (Thinking Machines Lab)
- Source: DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models
- Source: DeepSeek-V3/R1 Inference System Overview
- Source: DeepSeek-V3 Technical Report
- Source: DeepSeek-R1 (DeepSeek-AI, 2025)
- Source: DeepSeek V4 model cards (DeepSeek, Hugging Face)
- Source: DeepSeek API Docs — Thinking Mode
- Source: Dated Data: Tracing Knowledge Cutoffs in Large Language Models
- Source: Context windows (Claude API docs)
- Source: Context Rot: How Increasing Input Tokens Impacts LLM Performance (Chroma)
- Source: Claude's Extended Thinking (Anthropic, 2025)
- Source: Claude models overview (Anthropic)
- Source: Claude API Docs — Thinking
- Source: Claude API Docs — Steering Thinking
- Source: Claude API Docs — Extended Thinking (manual budget)
- Source: Claude API 2026-10-21 migration: sampling parameters removed
- Source: Choosing the right model (Anthropic)
- Source: Chat Completions API reference (OpenAI)
- Source: Chain-of-Verification Reduces Hallucination in Large Language Models (Meta, 2023)
- Source: Attention Is All You Need (Vaswani et al., 2017)
- Source: Anthropic, Introducing Computer Use (Claude 3.5 Sonnet)
- Source: Anthropic Vision Docs
- Source: Anthropic Tool Use Docs
- Source: Anthropic Structured Outputs Docs
- Source: Anthropic Streaming Messages Docs
- Source: Anthropic Prompt Caching Docs
- Source: Anthropic PDF Support Docs
- Source: Anthropic Message Batches Docs
- Source: Anthropic Computer Use Tool Docs
- Source: Anthropic Browser Use Tool Docs
- Source: A Survey on Hallucination in Large Language Models (Huang et al.)
- Source: 21 · Chain-of-Verification experiment (Notebook)
- Sampling parameters
- Reasoning Models
- Prompt Caching
- Open-Weight Models
- Multimodal Models
- Model Selection
- Mixture of Experts (MoE)
- Long-context degradation
- LLM Non-Determinism
- LLM limitations
- Hallucination
- Function Calling
- Frontier Model Landscape
- Extended Thinking
- Context window
- Computer-Use Agents
- Benchmarks
- Batch Processing