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": 76801.40685897361,
-      "required_pages_per_second_upper_estimate": 0.005313299530319402
+      "required_token_positions_per_second": 76247.62125502771,
+      "required_pages_per_second_upper_estimate": 0.00527498735727879
     },
     {
       "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": 9600.175857371702,
-      "required_pages_per_second_upper_estimate": 0.004019393157012918
+      "required_token_positions_per_second": 9530.952656878464,
+      "required_pages_per_second_upper_estimate": 0.003990410848511204
     },
     {
       "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": 19200.351714743403,
-      "required_pages_per_second_upper_estimate": 0.00399892691485595
+      "required_token_positions_per_second": 19061.905313756928,
+      "required_pages_per_second_upper_estimate": 0.003970092180607586
     },
     {
       "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": 28800.527572115105,
-      "required_pages_per_second_upper_estimate": 0.0017265476520889828
+      "required_token_positions_per_second": 28592.857970635396,
+      "required_pages_per_second_upper_estimate": 0.0017140981765733953
     },
     {
       "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": 28800.527572115105,
-      "required_pages_per_second_upper_estimate": 0.0022391143385991287
+      "required_token_positions_per_second": 28592.857970635396,
+      "required_pages_per_second_upper_estimate": 0.0022229689405261228
     },
     {
       "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": 28800.527572115105,
-      "required_pages_per_second_upper_estimate": 0.3694222440978836
+      "required_token_positions_per_second": 28592.857970635396,
+      "required_pages_per_second_upper_estimate": 0.36675848142834744
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.05997914969532303,
-    "producer_p95_seconds": 0.21315597399734543,
-    "gpu_update_seconds": 0.9599824145849678,
-    "producer_batches_per_second": 16.672460431328464,
-    "gpu_batches_per_second": 1.041685748412728,
-    "producer_headroom": 16.005268821939033,
+    "producer_batch_seconds": 0.08213303867205468,
+    "producer_p95_seconds": 0.2166471119999187,
+    "gpu_update_seconds": 0.9669547559182172,
+    "producer_batches_per_second": 12.175368355636458,
+    "gpu_batches_per_second": 1.0341745504425417,
+    "producer_headroom": 11.773030336538836,
     "sustainable": true,
-    "buffer_seconds": 60.47889211885297,
+    "buffer_seconds": 60.918149622847686,
     "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": 300706.3048313219,
-    "upload_bytes_per_second": 5587034.010911487,
+    "generation_bytes_per_second": 324070.903395807,
+    "upload_bytes_per_second": 5443844.699067282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.52457293228858,
+    "estimated_full_shards_per_hour": 69.53807188420923,
     "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": 1.3739831450002384,
-      "bytes": 13357834,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.1055001610002364,
       "bytes": 6014129,
@@ -562,6 +556,12 @@
       "bytes": 6571585,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1422953030014469,
+      "bytes": 877918,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -181,3 +181,4 @@
 {"step": 543100, "loss": 0.6443955898284912, "lr": 0.0003, "grad_norm": 0.21050887194058637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79492136196, "tokens_per_second": 154191.7396083871, "data_wait_s": 1.8157999875256792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 543200, "loss": 0.5875047445297241, "lr": 0.0003, "grad_norm": 0.19520319093406965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79506905431, "tokens_per_second": 154025.21101309868, "data_wait_s": 8.560000424040481e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 543300, "loss": 0.6209806203842163, "lr": 0.0003, "grad_norm": 0.17879850651620527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79521658934, "tokens_per_second": 153872.41254341637, "data_wait_s": 9.962997864931822e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 543400, "loss": 0.5429908633232117, "lr": 0.0003, "grad_norm": 0.15299706480539343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79536370222, "tokens_per_second": 154808.0225924273, "data_wait_s": 9.876002877717838e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa173d110181945a5411e1f29847c214a60246a58c5cb5596c0776546894f000
-size 390108
+oid sha256:8aea057577e9174531d96deca23f641595e60e782c58d1e0a7351e8a914c88bc
+size 392226