ui-svelte: fix cached tokens total counting -1 sentinel (#760)

The backend uses cache_tokens=-1 as a sentinel for endpoints that don't
report cache stats (embeddings, vLLM). The activity table correctly
renders these as "-", but the totals widget summed the sentinels
directly, so each such request subtracted 1 from the displayed total.

- clamp cache_tokens with Math.max(0, ...) when reducing

internal/perf/monitor_unix.go

@@ -139,7 +139,7 @@ func tryNvidiaSmi(ctx context.Context, every time.Duration, logger *logmon.Monit
 	cmd := exec.CommandContext(ctx, "nvidia-smi",
 		"--query-gpu=index,name,uuid,temperature.gpu,utilization.gpu,memory.used,memory.total,fan.speed,power.draw",
 		"--format=csv,noheader,nounits",
-		"-loop", fmt.Sprintf("%d", sec),
+		"--loop", fmt.Sprintf("%d", sec),
 	)
 
 	stdout, err := cmd.StdoutPipe()

ui-svelte/src/components/ActivityStats.svelte

@@ -11,7 +11,7 @@
     const totalRequests = $metrics.length;
     const totalInputTokens = $metrics.reduce((sum, m) => sum + m.tokens.input_tokens, 0);
     const totalOutputTokens = $metrics.reduce((sum, m) => sum + m.tokens.output_tokens, 0);
-    const totalCacheTokens = $metrics.reduce((sum, m) => sum + m.tokens.cache_tokens, 0);
+    const totalCacheTokens = $metrics.reduce((sum, m) => sum + Math.max(0, m.tokens.cache_tokens), 0);
 
     const promptPerSecond = $metrics.filter((m) => m.tokens.prompt_per_second > 0).map((m) => m.tokens.prompt_per_second);