Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76896.95381140515,
-      "required_pages_per_second_upper_estimate": 0.00531990969018809
+      "required_token_positions_per_second": 76327.82671256115,
+      "required_pages_per_second_upper_estimate": 0.005280536156933254
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9612.119226425644,
-      "required_pages_per_second_upper_estimate": 0.004024393596229896
+      "required_token_positions_per_second": 9540.978339070143,
+      "required_pages_per_second_upper_estimate": 0.003994608392284809
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19224.23845285129,
-      "required_pages_per_second_upper_estimate": 0.004003901892468177
+      "required_token_positions_per_second": 19081.956678140286,
+      "required_pages_per_second_upper_estimate": 0.003974268351018601
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28836.357679276935,
-      "required_pages_per_second_upper_estimate": 0.0017286956123039286
+      "required_token_positions_per_second": 28622.935017210428,
+      "required_pages_per_second_upper_estimate": 0.0017159012495905775
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28836.357679276935,
-      "required_pages_per_second_upper_estimate": 0.0022418999718309754
+      "required_token_positions_per_second": 28622.935017210428,
+      "required_pages_per_second_upper_estimate": 0.002225307298602385
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28836.357679276935,
-      "required_pages_per_second_upper_estimate": 0.36988183424118387
+      "required_token_positions_per_second": 28622.935017210428,
+      "required_pages_per_second_upper_estimate": 0.3671442774875954
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14908440150003344,
-    "producer_p95_seconds": 0.3460359509990667,
-    "gpu_update_seconds": 0.958789605383105,
-    "producer_batches_per_second": 6.70760985011417,
-    "gpu_batches_per_second": 1.0429816868951438,
-    "producer_headroom": 6.431186601254793,
+    "producer_batch_seconds": 0.0948122117892467,
+    "producer_p95_seconds": 0.3297427289944608,
+    "gpu_update_seconds": 0.9659386776155479,
+    "producer_batches_per_second": 10.547164559591224,
+    "gpu_batches_per_second": 1.0352624065831317,
+    "producer_headroom": 10.187914187285118,
     "sustainable": true,
-    "buffer_seconds": 60.403745139135616,
+    "buffer_seconds": 60.854136689779516,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 485854.199486078,
-    "upload_bytes_per_second": 3867452.8022042103,
+    "generation_bytes_per_second": 504705.5425618157,
+    "upload_bytes_per_second": 3866518.794878541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.25300110279804,
+    "estimated_full_shards_per_hour": 108.29806049004415,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.4411041519997525,
-      "bytes": 2727332,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.913498541998706,
       "bytes": 6177731,
@@ -562,6 +556,12 @@
       "bytes": 7184499,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0425019949980197,
+      "bytes": 1083508,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 836600, "loss": 0.5593646168708801, "lr": 3e-05, "grad_norm": 0.28687618863860886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122454269431, "tokens_per_second": 152319.09347932588, "data_wait_s": 1.5463992895092815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 836700, "loss": 0.48075222969055176, "lr": 3e-05, "grad_norm": 0.2749761565863887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122468915044, "tokens_per_second": 151792.26771400645, "data_wait_s": 1.4673998521175236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 836800, "loss": 0.45729824900627136, "lr": 3e-05, "grad_norm": 0.3163129596705552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122483637692, "tokens_per_second": 152265.28736622035, "data_wait_s": 1.0912001016549766e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 836900, "loss": 0.54319828748703, "lr": 3e-05, "grad_norm": 0.299937191501277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122498248230, "tokens_per_second": 151109.9856754473, "data_wait_s": 1.3634002243634313e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c2b2cd406bc2d0f41f561b93e9813fde7788a3104dc7894b7784f79862f5ea3
-size 785070
+oid sha256:4700588de62346073a6b2170823a9a480a533c569807c1adf9d581ae04c1f6a9
+size 787188