MCPcopy Create free account
hub / github.com/github/gh-aw / providerIncludesCacheReadsInInput

Function providerIncludesCacheReadsInInput

pkg/cli/effective_tokens.go:26–46  ·  view source on GitHub ↗

This file (effective_tokens.go) implements the Effective Tokens (ET) specification defined in docs/src/content/docs/specs/effective-tokens-specification.md. Effective Tokens normalize raw token counts across token classes and model pricing using the formula: base_weighted_tokens = (w_in × I) + (w

(normalizedProvider string)

Source from the content-addressed store, hash-verified

24// - Populating effective token counts on TokenUsageSummary after parsing
25
26func providerIncludesCacheReadsInInput(normalizedProvider string) bool {
27 // Cache read accounting is provider-specific:
28 // - bundled semantics: cache_read_tokens are already included in input_tokens,
29 // so we subtract once before applying input weight.
30 // - additive semantics: cache_read_tokens are separate from input_tokens,
31 // so no subtraction is applied.
32 //
33 // Known providers currently using bundled semantics are listed below.
34 // Unknown non-empty providers default to additive semantics to avoid
35 // under-counting input tokens. Empty provider values are treated as bundled
36 // semantics for backward compatibility with older usage records that omitted
37 // the provider field.
38 // We include both "azure-openai" and "azure_openai" to handle observed
39 // provider naming variants in historical logs.
40 switch normalizedProvider {
41 case "", "anthropic", "openai", "azure-openai", "azure_openai":
42 return true
43 default:
44 return false
45 }
46}

Callers 1

Calls

no outgoing calls

Tested by

no test coverage detected