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": 76225.68430360989,
-      "required_pages_per_second_upper_estimate": 0.0052734697080789655
+      "required_token_positions_per_second": 76565.6825054801,
+      "required_pages_per_second_upper_estimate": 0.005296991572588864
     },
     {
       "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": 9528.210537951236,
-      "required_pages_per_second_upper_estimate": 0.003989262780578328
+      "required_token_positions_per_second": 9570.710313185013,
+      "required_pages_per_second_upper_estimate": 0.004007056548972483
     },
     {
       "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": 19056.42107590247,
-      "required_pages_per_second_upper_estimate": 0.003968949958491581
+      "required_token_positions_per_second": 19141.420626370025,
+      "required_pages_per_second_upper_estimate": 0.003986653123265086
     },
     {
       "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": 28584.63161385371,
-      "required_pages_per_second_upper_estimate": 0.001713605019045253
+      "required_token_positions_per_second": 28712.130939555034,
+      "required_pages_per_second_upper_estimate": 0.0017212484089408492
     },
     {
       "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": 28584.63161385371,
-      "required_pages_per_second_upper_estimate": 0.0022223293774702663
+      "required_token_positions_per_second": 28712.130939555034,
+      "required_pages_per_second_upper_estimate": 0.002232241889233279
     },
     {
       "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": 28584.63161385371,
-      "required_pages_per_second_upper_estimate": 0.36665296255632573
+      "required_token_positions_per_second": 28712.130939555034,
+      "required_pages_per_second_upper_estimate": 0.3682883870083123
     }
   ],
   "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": 0.3624090036017833,
-    "producer_p95_seconds": 0.3660714050056413,
-    "gpu_update_seconds": 0.9672330353419785,
-    "producer_batches_per_second": 2.7593133450371026,
-    "gpu_batches_per_second": 1.0338770114964448,
-    "producer_headroom": 2.6688990221798647,
+    "producer_batch_seconds": 0.09322483208632093,
+    "producer_p95_seconds": 0.33173879900277825,
+    "gpu_update_seconds": 0.9629379323396354,
+    "producer_batches_per_second": 10.726755711118432,
+    "gpu_batches_per_second": 1.0384885322466375,
+    "producer_headroom": 10.329199965176759,
     "sustainable": true,
-    "buffer_seconds": 60.935681226544645,
+    "buffer_seconds": 60.66508973739703,
     "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": 334352.1699397766,
-    "upload_bytes_per_second": 4922072.837558684,
+    "generation_bytes_per_second": 333665.4330020599,
+    "upload_bytes_per_second": 4945616.549806968,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.74419234890912,
+    "estimated_full_shards_per_hour": 71.59683458849284,
     "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": 1.7213281089971133,
-      "bytes": 11058082,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.1525909759984643,
       "bytes": 12461215,
@@ -562,6 +556,12 @@
       "bytes": 2410876,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.1969935200031614,
+      "bytes": 10533402,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 962500, "loss": 0.5486325025558472, "lr": 3e-05, "grad_norm": 0.3615660195929653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140884058809, "tokens_per_second": 150130.3312679302, "data_wait_s": 1.655399682931602e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962600, "loss": 0.5889420509338379, "lr": 3e-05, "grad_norm": 0.36584894136504326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140898711932, "tokens_per_second": 152280.7069078466, "data_wait_s": 1.4067001757211983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962700, "loss": 0.5180770754814148, "lr": 3e-05, "grad_norm": 0.34582609136541403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140913428967, "tokens_per_second": 152893.97975429203, "data_wait_s": 6.0377002228051424e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 962800, "loss": 0.514863908290863, "lr": 3e-05, "grad_norm": 0.34611537754554306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140928191675, "tokens_per_second": 153670.39868493073, "data_wait_s": 1.0747004125732929e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1e23acb29ab4dfe4efdb47de752c4546e65ae0c850458d85e883cfc323b8ca2
-size 804132
+oid sha256:e342386c5ceb84539e26b01ec997388f108ea43fdc475b0d99bebc57c4d8f1c9
+size 806250