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": 77160.6514700967,
-      "required_pages_per_second_upper_estimate": 0.0053381529060792365
+      "required_token_positions_per_second": 77167.28713320478,
+      "required_pages_per_second_upper_estimate": 0.005338611976649914
     },
     {
       "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": 9645.081433762087,
-      "required_pages_per_second_upper_estimate": 0.004038194184112503
+      "required_token_positions_per_second": 9645.910891650597,
+      "required_pages_per_second_upper_estimate": 0.004038541460809373
     },
     {
       "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": 19290.162867524174,
-      "required_pages_per_second_upper_estimate": 0.004017632209500813
+      "required_token_positions_per_second": 19291.821783301195,
+      "required_pages_per_second_upper_estimate": 0.004017977717908615
     },
     {
       "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": 28935.244301286257,
-      "required_pages_per_second_upper_estimate": 0.0017346237143022584
+      "required_token_positions_per_second": 28937.732674951792,
+      "required_pages_per_second_upper_estimate": 0.0017347728885039806
     },
     {
       "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": 28935.244301286257,
-      "required_pages_per_second_upper_estimate": 0.0022495879717359174
+      "required_token_positions_per_second": 28937.732674951792,
+      "required_pages_per_second_upper_estimate": 0.0022497814318432137
     },
     {
       "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": 28935.244301286257,
-      "required_pages_per_second_upper_estimate": 0.37115024565213706
+      "required_token_positions_per_second": 28937.732674951792,
+      "required_pages_per_second_upper_estimate": 0.3711821638377111
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.22211623922648727,
-    "producer_p95_seconds": 0.16718530399975862,
-    "gpu_update_seconds": 0.9555129278369688,
-    "producer_batches_per_second": 4.5021471797040515,
-    "gpu_batches_per_second": 1.0465583152953652,
-    "producer_headroom": 4.30185983323197,
+    "producer_batch_seconds": 0.08582025112495728,
+    "producer_p95_seconds": 0.16327000799992675,
+    "gpu_update_seconds": 0.9554307626848674,
+    "producer_batches_per_second": 11.652261405573903,
+    "gpu_batches_per_second": 1.046648317236393,
+    "producer_headroom": 11.132929001730918,
     "sustainable": true,
-    "buffer_seconds": 60.197314453729035,
+    "buffer_seconds": 60.19213804914664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471718.0651086896,
-    "upload_bytes_per_second": 5138876.858680128,
+    "generation_bytes_per_second": 465184.8884361477,
+    "upload_bytes_per_second": 4704068.82613897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.21971573777691,
+    "estimated_full_shards_per_hour": 99.81784810841869,
     "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": [
@@ -244,6 +244,18 @@
       "bytes": 12097526,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.8632086769994203,
+      "bytes": 2602865,
+      "pages": 34,
+      "shards": 2
+    },
+    {
+      "seconds": 1.1280111480000414,
+      "bytes": 298657,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 764800, "loss": 0.5302116870880127, "lr": 3e-05, "grad_norm": 0.22984702001245264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111944009062, "tokens_per_second": 152285.7056770709, "data_wait_s": 1.1764000191760715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 764900, "loss": 0.5336251258850098, "lr": 3e-05, "grad_norm": 0.2608528531899802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111958687911, "tokens_per_second": 152109.9223911244, "data_wait_s": 1.7841000044427346e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765000, "loss": 0.5579110383987427, "lr": 3e-05, "grad_norm": 0.2431269280429593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111973474399, "tokens_per_second": 153445.67202687074, "data_wait_s": 2.4105999727908056e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 765100, "loss": 0.5478135943412781, "lr": 3e-05, "grad_norm": 0.24365099445058058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111988119969, "tokens_per_second": 141452.28978523877, "data_wait_s": 1.1377000191714615e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d47c9355a4b9383b5fc8e0aff985fd48780d531cc6d8cc0ba20bb59e79ac596c
-size 106848
+oid sha256:3cb4ea7321122fa3f3089efb8f8f27159f4be743af83fc7edc04d1e296c806e5
+size 108966