kreuzberg 4.9.7

High-performance document intelligence library for Rust. Extract text, metadata, and structured data from PDFs, Office documents, images, and 91+ formats and 248 programming languages via tree-sitter code intelligence with async/sync APIs.
Documentation
//! Helpers for extracting LLM usage metadata from liter-llm responses.

use crate::types::LlmUsage;

/// Extract usage metadata from a chat completion response.
pub(crate) fn extract_usage_from_chat(response: &liter_llm::ChatCompletionResponse, source: &str) -> Option<LlmUsage> {
    Some(LlmUsage {
        model: response.model.clone(),
        source: source.to_string(),
        input_tokens: response.usage.as_ref().map(|u| u.prompt_tokens),
        output_tokens: response.usage.as_ref().map(|u| u.completion_tokens),
        total_tokens: response.usage.as_ref().map(|u| u.total_tokens),
        estimated_cost: response.estimated_cost(),
        finish_reason: response
            .choices
            .first()
            .and_then(|c| c.finish_reason.as_ref())
            .map(|fr| format!("{fr:?}").to_lowercase()),
    })
}

/// Extract usage metadata from an embedding response.
pub(crate) fn extract_usage_from_embedding(response: &liter_llm::EmbeddingResponse, source: &str) -> Option<LlmUsage> {
    Some(LlmUsage {
        model: response.model.clone(),
        source: source.to_string(),
        input_tokens: response.usage.as_ref().map(|u| u.prompt_tokens),
        output_tokens: response.usage.as_ref().map(|u| u.completion_tokens),
        total_tokens: response.usage.as_ref().map(|u| u.total_tokens),
        estimated_cost: response.estimated_cost(),
        finish_reason: None,
    })
}

/// Append an `LlmUsage` entry to an `ExtractionResult`, lazily initializing the vec.
pub(crate) fn push_llm_usage(result: &mut crate::ExtractionResult, usage: Option<LlmUsage>) {
    if let Some(u) = usage {
        result.llm_usage.get_or_insert_with(Vec::new).push(u);
    }
}