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": 73791.83096575811,
-      "required_pages_per_second_upper_estimate": 0.005105090086848587
+      "required_token_positions_per_second": 77187.65313174498,
+      "required_pages_per_second_upper_estimate": 0.005340020943684552
     },
     {
       "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": 9223.978870719764,
-      "required_pages_per_second_upper_estimate": 0.0038618873345881554
+      "required_token_positions_per_second": 9648.456641468123,
+      "required_pages_per_second_upper_estimate": 0.004039607313096669
     },
     {
       "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": 18447.957741439528,
-      "required_pages_per_second_upper_estimate": 0.0038422230921801434
+      "required_token_positions_per_second": 19296.913282936246,
+      "required_pages_per_second_upper_estimate": 0.004019038143010721
     },
     {
       "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": 27671.93661215929,
-      "required_pages_per_second_upper_estimate": 0.0016588903472982476
+      "required_token_positions_per_second": 28945.36992440437,
+      "required_pages_per_second_upper_estimate": 0.001735230729947258
     },
     {
       "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": 27671.93661215929,
-      "required_pages_per_second_upper_estimate": 0.00215137135560957
+      "required_token_positions_per_second": 28945.36992440437,
+      "required_pages_per_second_upper_estimate": 0.002250375194395442
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 27671.93661215929,
-      "required_pages_per_second_upper_estimate": 0.35494589105013136
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28945.36992440437,
+      "required_pages_per_second_upper_estimate": 0.3712801262734492
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2562649832969157,
-    "producer_p95_seconds": 0.3117504590009048,
-    "gpu_update_seconds": 0.9991349860150817,
-    "producer_batches_per_second": 3.9022108566482228,
-    "gpu_batches_per_second": 1.0008657628819189,
-    "producer_headroom": 3.898835389685122,
+    "producer_batch_seconds": 0.08750313167954005,
+    "producer_p95_seconds": 0.320557426999585,
+    "gpu_update_seconds": 0.9551786718292885,
+    "producer_batches_per_second": 11.42816240751552,
+    "gpu_batches_per_second": 1.0469245487704126,
+    "producer_headroom": 10.915936989860079,
     "sustainable": true,
-    "buffer_seconds": 60.94723414691998,
+    "buffer_seconds": 60.17625632524518,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460345.019986264,
+    "generation_bytes_per_second": 462874.1932209936,
     "upload_bytes_per_second": 4185822.474152754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.7793250054449,
+    "estimated_full_shards_per_hour": 99.32202670547825,
     "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": [

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -324,3 +324,4 @@
 {"step": 874900, "loss": 0.537115216255188, "lr": 3e-05, "grad_norm": 0.30592999147303934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128060867543, "tokens_per_second": 154272.42569300413, "data_wait_s": 8.540002454537898e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 875000, "loss": 0.5282732248306274, "lr": 3e-05, "grad_norm": 0.30791818889206707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128075402216, "tokens_per_second": 150694.60485398571, "data_wait_s": 1.2407999747665599e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875100, "loss": 0.4834609925746918, "lr": 3e-05, "grad_norm": 0.3024347114563917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128090022645, "tokens_per_second": 139551.35050221166, "data_wait_s": 1.2865999451605603e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 875200, "loss": 0.5135159492492676, "lr": 3e-05, "grad_norm": 0.3100302680594288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128104481050, "tokens_per_second": 149996.80240268543, "data_wait_s": 8.998002158477902e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a92cdca6dcde36b724284863843d2f7108e8d6162c812972e78474534c19c22
-size 696114
+oid sha256:964457544950eaf9149157e9f110521e704af4e1639a822a3a5455fddb814f57
+size 698232