Expose cache usage in adapters
This commit is contained in:
@@ -86,6 +86,8 @@ type tokenUsageDTO struct {
|
||||
PromptTokens int `json:"prompt_tokens"`
|
||||
CompletionTokens int `json:"completion_tokens"`
|
||||
TotalTokens int `json:"total_tokens"`
|
||||
CachedTokens int `json:"cached_tokens"`
|
||||
CacheWriteTokens int `json:"cache_write_tokens"`
|
||||
}
|
||||
|
||||
type validationDTO struct {
|
||||
|
||||
@@ -105,6 +105,8 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
|
||||
PromptTokens: res.Usage.PromptTokens,
|
||||
CompletionTokens: res.Usage.CompletionTokens,
|
||||
TotalTokens: res.Usage.TotalTokens,
|
||||
CachedTokens: res.Usage.CachedTokens,
|
||||
CacheWriteTokens: res.Usage.CacheWriteTokens,
|
||||
},
|
||||
StartTime: res.StartTime,
|
||||
EndTime: res.EndTime,
|
||||
|
||||
@@ -66,11 +66,17 @@ func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
|
||||
APIKeyEnv: envName,
|
||||
},
|
||||
InputHashes: map[string]string{"transcript": "h1"},
|
||||
Usage: domain.TokenUsage{PromptTokens: 1, CompletionTokens: 2, TotalTokens: 3},
|
||||
StartTime: start,
|
||||
EndTime: end,
|
||||
Duration: 2 * time.Second,
|
||||
RawOutput: "hello",
|
||||
Usage: domain.TokenUsage{
|
||||
PromptTokens: 1,
|
||||
CompletionTokens: 2,
|
||||
TotalTokens: 3,
|
||||
CachedTokens: 4,
|
||||
CacheWriteTokens: 5,
|
||||
},
|
||||
StartTime: start,
|
||||
EndTime: end,
|
||||
Duration: 2 * time.Second,
|
||||
RawOutput: "hello",
|
||||
}}
|
||||
|
||||
h := NewHandler(r)
|
||||
@@ -111,6 +117,13 @@ func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
|
||||
if metadata["model_name"] != "m1" || metadata["endpoint"] != "http://llm/v1" {
|
||||
t.Fatalf("unexpected model metadata: name=%#v endpoint=%#v", metadata["model_name"], metadata["endpoint"])
|
||||
}
|
||||
usage := metadata["usage"].(map[string]any)
|
||||
if usage["prompt_tokens"] != float64(1) || usage["completion_tokens"] != float64(2) || usage["total_tokens"] != float64(3) {
|
||||
t.Fatalf("unexpected base usage metadata: %#v", usage)
|
||||
}
|
||||
if usage["cached_tokens"] != float64(4) || usage["cache_write_tokens"] != float64(5) {
|
||||
t.Fatalf("unexpected cache usage metadata: %#v", usage)
|
||||
}
|
||||
modelParams := metadata["model_params"].(map[string]any)
|
||||
if modelParams["api_key_env"] != envName {
|
||||
t.Fatalf("expected model_params.api_key_env=%q, got %#v", envName, modelParams["api_key_env"])
|
||||
@@ -171,6 +184,10 @@ func TestHandlerPostRunsSuccessUsingPromptDefaultProfile(t *testing.T) {
|
||||
if metadata["selected_profile_id"] != "prompt-default" {
|
||||
t.Fatalf("expected selected_profile_id from result, got %#v", metadata["selected_profile_id"])
|
||||
}
|
||||
usage := metadata["usage"].(map[string]any)
|
||||
if usage["cached_tokens"] != float64(0) || usage["cache_write_tokens"] != float64(0) {
|
||||
t.Fatalf("expected zero cache usage fields to be included, got %#v", usage)
|
||||
}
|
||||
}
|
||||
|
||||
func TestHandlerModelOverrideMapsAllSupportedExecutionFields(t *testing.T) {
|
||||
|
||||
Reference in New Issue
Block a user