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": 77519.91678204635,
-      "required_pages_per_second_upper_estimate": 0.005363007713970287
+      "required_token_positions_per_second": 77490.77832362556,
+      "required_pages_per_second_upper_estimate": 0.005360991847806198
     },
     {
       "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": 9689.989597755794,
-      "required_pages_per_second_upper_estimate": 0.004056996294587565
+      "required_token_positions_per_second": 9686.347290453195,
+      "required_pages_per_second_upper_estimate": 0.004055471336579993
     },
     {
       "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": 19379.979195511587,
-      "required_pages_per_second_upper_estimate": 0.004036338582004739
+      "required_token_positions_per_second": 19372.69458090639,
+      "required_pages_per_second_upper_estimate": 0.004034821388890683
     },
     {
       "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": 29069.96879326738,
-      "required_pages_per_second_upper_estimate": 0.0017427002418841386
+      "required_token_positions_per_second": 29059.041871359583,
+      "required_pages_per_second_upper_estimate": 0.0017420451896002099
     },
     {
       "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": 29069.96879326738,
-      "required_pages_per_second_upper_estimate": 0.0022600622083970364
+      "required_token_positions_per_second": 29059.041871359583,
+      "required_pages_per_second_upper_estimate": 0.002259212688281155
     },
     {
       "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": 29069.96879326738,
-      "required_pages_per_second_upper_estimate": 0.37287834677938175
+      "required_token_positions_per_second": 29059.041871359583,
+      "required_pages_per_second_upper_estimate": 0.3727381879575632
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.21814840160945437,
-    "producer_p95_seconds": 0.23167979199934052,
-    "gpu_update_seconds": 0.9510846123234673,
-    "producer_batches_per_second": 4.584035420943744,
-    "gpu_batches_per_second": 1.0514311629509325,
-    "producer_headroom": 4.359805551205323,
+    "producer_batch_seconds": 0.07432439335161689,
+    "producer_p95_seconds": 0.20063424699947063,
+    "gpu_update_seconds": 0.951442243773691,
+    "producer_batches_per_second": 13.454532958905684,
+    "gpu_batches_per_second": 1.0510359473147999,
+    "producer_headroom": 12.801211027348304,
     "sustainable": true,
-    "buffer_seconds": 60.86941518870191,
+    "buffer_seconds": 60.892303601516225,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392329.06402836315,
-    "upload_bytes_per_second": 5917448.422021412,
+    "generation_bytes_per_second": 390753.5515383779,
+    "upload_bytes_per_second": 5889673.536274945,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.18468418729945,
+    "estimated_full_shards_per_hour": 83.8466158830023,
     "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": 0.9458256010002515,
-      "bytes": 5937789,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.3173817269998835,
       "bytes": 11936639,
@@ -562,6 +556,12 @@
       "bytes": 7119189,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.3181563210000604,
+      "bytes": 6116642,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -112,3 +112,4 @@
 {"step": 496200, "loss": 0.5768790245056152, "lr": 0.0003, "grad_norm": 0.1476545211548444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72627194811, "tokens_per_second": 150266.67203299157, "data_wait_s": 8.623999747214839e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 496300, "loss": 0.5547523498535156, "lr": 0.0003, "grad_norm": 0.1692655133963257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72641635228, "tokens_per_second": 151173.60173753466, "data_wait_s": 1.1692000043694861e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 496400, "loss": 0.5905017852783203, "lr": 0.0003, "grad_norm": 0.15937914559476943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72656150286, "tokens_per_second": 151870.15509323424, "data_wait_s": 1.6418000086559914e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 496500, "loss": 0.6291759610176086, "lr": 0.0003, "grad_norm": 0.2717890904557022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72670760815, "tokens_per_second": 152419.8958035513, "data_wait_s": 1.2229000276420265e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9b5d1c393379eb5583470d6e50bc07c90323084ab8a8921392bab0d8c07be8c
-size 243183
+oid sha256:52f8293d99c91d7adac1a89b73590de74882f331527a7efc546e1d23e63d64c2
+size 245301