Demonstrate with a real-world example

演示搜索功能

Demonstrate the improved single test case selection

Demonstrate the improved single test case selection

Demonstrate the contextual retrieval solution

Demonstrate system hint features like timestamps, tool counting, and TODO tracking

Demonstrate tool call loop prevention

Demo with unified ToolCallingAgent that auto-selects backend

Demonstrate the improved test cases table with index numbers

Demonstrate the improved test cases table with index numbers

Demo streaming with vLLM backend

Demonstrate weather checking with learning.

Demonstrate web search capability

Demonstrate how the agent logs tool calls and results

Demonstrate how the agent logs tool calls and results

Demonstrate the attention tracking functionality

Demonstrate the conversation-based memory processing

Show what the evaluation output looks like

Show what the evaluation output looks like

Demonstrate processing with different conversation intervals

Demonstrate the ReAct agent with various queries

Demonstrate how streaming would work (matching Go handleStreamingResponse)

Demonstrate the tool calling functionality

Detect Level 3 PII in conversation text using local LLM Args: conversation_text: Text to analyze

Display detailed information about a test case.

Download Chatbot Arena voting data via HTTPS. Args: output_path: Path to save downloaded file force_download: If True, r

Download a file from a URL. Args: url: URL to download from output_path: Local path to save the file overwrite:

Download Qwen3-0.6B model from ModelScope This is optional - vLLM can download from HuggingFace automatically

Download ArXiv paper PDF. Args: paper_id: ArXiv paper ID download_dir: Directory to save PDF Returns:

Download YouTube video using yt-dlp. Args: url: YouTube video URL output_dir: Directory to save video max_resolu

Encode multiple texts into embeddings. Args: texts: List of input texts to encode return_sparse: Whe

* Ensure temp directory exists

Evaluate the trained agent without learning. Args: num_episodes: Number of episodes to evaluate verb

Evaluate all loaded test cases

Evaluate all test cases in a category

从 parquet 文件评测

Evaluate the distilled model on test sentences.

Evaluate a single response

Evaluate a single response

Evaluate a single test case

Evaluate a specific test case selected by the user

Run evaluation on all test cases in a category

Evaluate multiple test cases. Args: test_cases: List of test cases to evaluate agent_responses: Dict

Example: Code interpreter with analysis.

Example: Custom task from user input.

Example: File operations with verification.

Example: Search on Google.

Example: Virtual terminal.

Example: Check weather.

Execute the chain and return all results Returns: List of all chain results

Execute the ReAct loop for a given query Args: query: User query to answer temperature: Sampling tem

Execute a single task with the agent

Execute a tool by name with given arguments

Fast expected score calculation using Numba JIT. Args: rating_a: Rating of model A rating_b: Rating of model B

Export test cases to different formats.

Export test cases to JSON format.

Export test cases to Markdown format.

从模型输出中提取并标准化答案

从 \\boxed{} 格式中提取答案（同时支持 \\(\\boxed{}\\) 形式）

Extract audio file metadata using ffprobe. Args: file_path: Path to audio file Returns: TextContent with au

Extract entities and relationships from text using LLM.

Extract text from image using OCR. Args: image_path: Path to image file language: OCR language (eng, chi_sim, etc.)

Extract keyframes from video. Args: video_path: Path to video file num_frames: Number of keyframes to extract

事实核查 - 验证陈述的真实性 Args: statement: 需要验证的陈述 Returns: 验证结果

Three learning paradigms — caption 图 1-1.

Context ablation experiment design — caption 图 1-2.

Agent trajectory — caption 图 1-3.

Kimi K2 / GPT-5 native agent architecture — caption 图 1-4

Autonomous Agent loop — caption 图 1-5.

Prompt chaining — workflow pattern (ch1 编排模式节).

Evaluator-optimizer — workflow pattern (ch1 编排模式节).

Orchestrator-workers — workflow pattern (ch1 编排模式节).

Parallelization — workflow pattern (ch1 编排模式节).

Routing — workflow pattern (ch1 编排模式节).

Context window with actual content snippets in each layer.

Qwen3-0.6B on local hardware + tool registry + ReAct loop.

Chat template token structure with actual token content and larger fonts.

KV Cache with concrete token sequences showing prefix reuse.

Show WHERE hints are inserted with actual hint text.

Data visualization comparing 6 strategies with actual experiment numbers.

6 compression strategies as pipeline variants.

Agent Skills with concrete PPTX example showing 3 layers.

4 memory modes showing how the same info is stored differently.

Find all workflows that might match the given task. Args: task_description: Natural language description of the

Format conversation dictionary into text

Format memory operations for display Args: operations: List of memory operations verbose: Whether to show detailed outpu

Format operation summary statistics Args: summary: Dictionary with counts of operations Returns: Formatted

Convert ISO timestamp to readable format

Create from dictionary

Create from dictionary

Create from dictionary