anthropic: add InputTokens to streaming response (#13934)

---------

Co-authored-by: ParthSareen <parth.sareen@ollama.com>

anthropic/anthropic.go

@@ -211,6 +211,7 @@ type MessageDelta struct {
 
 // DeltaUsage contains cumulative token usage
 type DeltaUsage struct {
+	InputTokens  int `json:"input_tokens"`
 	OutputTokens int `json:"output_tokens"`
 }
 
@@ -721,6 +722,7 @@ func (c *StreamConverter) Process(r api.ChatResponse) []StreamEvent {
 			})
 		}
 
+		c.inputTokens = r.Metrics.PromptEvalCount
 		c.outputTokens = r.Metrics.EvalCount
 		stopReason := mapStopReason(r.DoneReason, len(c.toolCallsSent) > 0)
 
@@ -732,6 +734,7 @@ func (c *StreamConverter) Process(r api.ChatResponse) []StreamEvent {
 					StopReason: stopReason,
 				},
 				Usage: DeltaUsage{
+					InputTokens:  c.inputTokens,
 					OutputTokens: c.outputTokens,
 				},
 			},

anthropic/anthropic_test.go

@@ -642,7 +642,7 @@ func TestStreamConverter_Basic(t *testing.T) {
 		},
 		Done:       true,
 		DoneReason: "stop",
-		Metrics:    api.Metrics{EvalCount: 5},
+		Metrics:    api.Metrics{PromptEvalCount: 10, EvalCount: 5},
 	}
 
 	events2 := conv.Process(resp2)
@@ -650,6 +650,24 @@ func TestStreamConverter_Basic(t *testing.T) {
 	// Should have content_block_delta, content_block_stop, message_delta, message_stop
 	hasStop := false
 	for _, e := range events2 {
+		if e.Event == "message_delta" {
+			if data, ok := e.Data.(MessageDeltaEvent); ok {
+				if data.Type != "message_delta" {
+					t.Errorf("unexpected data type: %+v", data)
+				}
+
+				if data.Delta.StopReason != "end_turn" {
+					t.Errorf("unexpected stop reason: %+v", data.Delta.StopReason)
+				}
+
+				if data.Usage.InputTokens != 10 || data.Usage.OutputTokens != 5 {
+					t.Errorf("unexpected usage: %+v", data.Usage)
+				}
+			} else {
+				t.Errorf("unexpected data: %+v", e.Data)
+			}
+		}
+
 		if e.Event == "message_stop" {
 			hasStop = true
 		}