Get current memory context as a string

Get mock search results for testing without API key Args: query: Search query Returns:

Get the number of transformer layers in the model

Get recent commits

Get recent commits

Get non-streaming response

Get system hint content with current state

Get system hint content with current state

Get system prompt based on mode

获取系统提示

Generate the system prompt

Generate the system prompt

Generate the system prompt

Generate the system prompt

Get current system state information

Generate a hash for a question to use as key.

Get top N most frequent terms across all documents

Get tool definitions based on mode

定义可用的工具 根据 Kimi 文档，$web_search 是内置工具

Get tool descriptions for the model

Get tool descriptions for the model

Get tool descriptions for the model

Get tool descriptions for the model

Get user profile message for dynamic profile mode

Apply windowed compression strategy to message history Only compresses when context usage exceeds 80% threshold Args

Index chunks to the knowledge base

Index chunks to the knowledge base

Index chunks to the retrieval pipeline

Index to Dify knowledge base

Index to Dify knowledge base

Index to local retrieval pipeline

Index to local retrieval pipeline

Infer the approach based on action types. Args: action_types: Count of action types tools_used: List

Initialize the LLM client based on provider

Initialize LLM client for context generation

Initialize the LLM client based on provider

Initialize LLM client for context generation

Initialize the LLM client based on provider

Initialize the LLM client based on provider

Initialize vLLM backend

Initialize the selected backend

Initialize the BGE-M3 model.

Create the game world with rooms and items - simplified for better learning.

Check if string is base64 encoded.

Simple similarity check between questions. Can be enhanced with semantic similarity using embeddings. Args:

Determine if a task execution was successful. Args: response: The task response task: The original t

Simple keyword-based search fallback. Args: query: The search query top_k: Number of results to retu

Learn from a successful task execution. Args: question: The original question response: The successf

Load existing experiences from file.

Load existing index from disk.

Load BM25 indexes from disk

Load a single test case from a YAML file.

Load a single test case from YAML file

Load a single test case from YAML file

Load system prompt from system-prompt.md

Load system prompt from system-prompt.md

Load all test cases from YAML files.

Load timer state from storage.

Load tool definitions from tools.json

Load tool definitions from tools.json

Log full request and response when in verbose mode Args: request_data: The request payload sent to the API

Move to another room.

Strategy 1: No compression - return all original content

Strategy 2B: Non-context-aware summarization - Concatenate all pages then summarize once

Strategy 2A: Non-context-aware summarization - Summarize each page individually then concatenate

Notify admin about pending approval request.

Output memory operations in a formatted way Args: results: Processing results with operations

Parse datetime string in various formats.

Parse the LLM's evaluation response.

Parse LLM response into structured summary. Args: response: LLM response object Returns:

Parse tool calls from model output Extracts content between <tool_call> tags

Prepare assistant message for adding to messages list, filtering out reasoning_content if in NO_REASONING mode Args

Prepare NumPy arrays from DataFrame for fast processing. Args: df: DataFrame with columns 'model_a', 'model_b',

Prepare input tensors for the model. Args: prompts: List of text prompts to convert to speech voice:

Generate memory cards for a test case For Layer 3 tests, this creates cards that enable proactive service

Print analysis of the results

Print comparison table of all strategies

Print summary for a single mode

Print summary for a single mode

Print processing statistics

Print processing statistics

Print a summary of the metrics

Process attention weights from generation output

Process using native multimodal capabilities

Process using Doubao's native multimodal API

Process using OpenAI's native multimodal API

Process PDF using pdfplumber for better structure preservation.

Fallback PDF processing using pypdf.

Combine BM25 and embedding results using reciprocal rank fusion. Educational Note: Rank fusion combines different re

Enforce rate limiting to comply with PubChem usage policy.

Read image file

Read Jupyter notebook

Read PDF file