Add debug stats

internal/llm/llm.go

@@ -59,11 +59,41 @@ type ChatRequest struct {
 	ServiceTier         string          `json:"service_tier,omitempty"`
 }
 
+// Usage represents token usage from the API response.
+type Usage struct {
+	PromptTokens            int           `json:"prompt_tokens"`
+	CompletionTokens        int           `json:"completion_tokens"`
+	TotalTokens             int           `json:"total_tokens"`
+	PromptTokensDetails     *TokenDetails `json:"prompt_tokens_details,omitempty"`
+	CompletionTokensDetails *TokenDetails `json:"completion_tokens_details,omitempty"`
+}
+
+// TokenDetails breaks down token usage by category.
+type TokenDetails struct {
+	ReasoningTokens int `json:"reasoning_tokens,omitempty"`
+	CachedTokens    int `json:"cached_tokens,omitempty"`
+}
+
+// Timings is a llama.cpp-specific extension to the chat completion
+// response. It is not part of the OpenAI API spec.
+type Timings struct {
+	PromptN             int     `json:"prompt_n"`
+	PromptMS            float64 `json:"prompt_ms"`
+	PromptPerTokenMS    float64 `json:"prompt_per_token_ms"`
+	PromptPerSecond     float64 `json:"prompt_per_second"`
+	PredictedN          int     `json:"predicted_n"`
+	PredictedMS         float64 `json:"predicted_ms"`
+	PredictedPerTokenMS float64 `json:"predicted_per_token_ms"`
+	PredictedPerSecond  float64 `json:"predicted_per_second"`
+}
+
 // ChatResponse is the non-streaming response from the API.
 type ChatResponse struct {
 	ID      string   `json:"id"`
 	Model   string   `json:"model"`
 	Choices []Choice `json:"choices"`
+	Usage   *Usage   `json:"usage,omitempty"`
+	Timings *Timings `json:"timings,omitempty"`
 }
 
 // Choice is a single choice in a ChatResponse.
@@ -246,6 +276,8 @@ type StreamChunk struct {
 		Delta        Delta  `json:"delta"`
 		FinishReason string `json:"finish_reason,omitempty"`
 	} `json:"choices"`
+	Usage   *Usage   `json:"usage,omitempty"`
+	Timings *Timings `json:"timings,omitempty"`
 }
 
 // StreamOptions controls streaming behavior.