Piloto concluído (tarefas 52–58): reports/pilot, viabilidade em CPU e parâmetros mantidos

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

reports/pilot/summary-laya-cuda.json

@@ -3,7 +3,7 @@
     "name": "laya",
     "model_id": "laya@1720e3e3357c",
     "revision": "1720e3e3357cfe1e281542e223f8273b0890ca34",
-    "adapter_version": "0.1.0",
+    "adapter_version": "0.2.0",
     "device": "cuda",
     "context_limit": 1024,
     "probabilities_source": "native"
@@ -17,27 +17,29 @@
   "batch_size": 1,
   "state_budget_tokens": 512,
   "process_guard": "ok",
-  "load_seconds": 8.46,
+  "load_seconds": 9.06,
   "warmup_latency_ms": [
-    260.7,
-    22.5,
-    27.4,
-    21.5,
-    21.8
+    264.4,
+    22.4,
+    28.5,
+    22.4,
+    22.5
   ],
   "latency_ms": {
     "n": 45,
-    "p50": 21.237645000013572,
-    "p95": 23.064290799999295,
-    "p99": 25.273738600021716,
-    "mean": 21.341669533332052
+    "p50": 22.042784999996456,
+    "p95": 22.97260880000067,
+    "p99": 23.23121632001289,
+    "mean": 21.861335755556233
   },
   "errors": {},
   "truncated": 0,
+  "with_diagnostics": 0,
+  "operational_env": {},
   "state_tokens_max": 88,
-  "rss_peak_mib": 2779.8515625,
+  "rss_peak_mib": 2773.41015625,
   "gpu_mem_peak_mib": 1811.0,
-  "gpu_util_mean_pct": 6.785714285714286,
+  "gpu_util_mean_pct": 6.857142857142857,
   "quality": {
     "n": 50,
     "errors": 0,
@@ -53,6 +55,6 @@
   },
   "oci": {
     "region": "sa-saopaulo-1",
-    "run_id": "20261007T174356Z-laya-evaluate"
+    "run_id": "20261007T180950Z-laya-evaluate"
   }
 }