Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74576.43937480818,
-      "required_pages_per_second_upper_estimate": 0.006464096752750811
+      "required_token_positions_per_second": 74812.39898306069,
+      "required_pages_per_second_upper_estimate": 0.006484549133559979
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9322.054921851022,
-      "required_pages_per_second_upper_estimate": 0.003718707085467936
+      "required_token_positions_per_second": 9351.549872882586,
+      "required_pages_per_second_upper_estimate": 0.003730473062423243
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18644.109843702045,
-      "required_pages_per_second_upper_estimate": 0.004312986687214084
+      "required_token_positions_per_second": 18703.099745765172,
+      "required_pages_per_second_upper_estimate": 0.00432663296286957
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27966.16476555307,
-      "required_pages_per_second_upper_estimate": 0.001702992707095327
+      "required_token_positions_per_second": 28054.649618647756,
+      "required_pages_per_second_upper_estimate": 0.001708380970404648
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27966.16476555307,
-      "required_pages_per_second_upper_estimate": 0.0021529127789511457
+      "required_token_positions_per_second": 28054.649618647756,
+      "required_pages_per_second_upper_estimate": 0.002159724587884125
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27966.16476555307,
-      "required_pages_per_second_upper_estimate": 0.3472461696556001
+      "required_token_positions_per_second": 28054.649618647756,
+      "required_pages_per_second_upper_estimate": 0.34834485539128296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09813341547657828,
-    "producer_p95_seconds": 0.1661170279999169,
-    "gpu_update_seconds": 0.9886232249498521,
-    "producer_batches_per_second": 10.190208861513357,
-    "gpu_batches_per_second": 1.0115076955133488,
-    "producer_headroom": 10.074277147581896,
+    "producer_batch_seconds": 0.4706628974296869,
+    "producer_p95_seconds": 0.1725796470000205,
+    "gpu_update_seconds": 0.9855050900946751,
+    "producer_batches_per_second": 2.1246629072762033,
+    "gpu_batches_per_second": 1.0147081025263223,
+    "producer_headroom": 2.093866109856049,
     "sustainable": true,
-    "buffer_seconds": 60.30601672194098,
+    "buffer_seconds": 60.115810495775186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295546.3863260697,
-    "upload_bytes_per_second": 4938498.620481795,
+    "generation_bytes_per_second": 263785.65107051184,
+    "upload_bytes_per_second": 4932607.328702079,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.4173745378167,
+    "estimated_full_shards_per_hour": 56.60226010405079,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 7901550,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.067005848000008,
+      "bytes": 5080800,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -29,3 +29,4 @@
 {"step": 317900, "loss": 0.5926313735544682, "lr": 0.0003, "grad_norm": 0.16847278284709633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46531859727, "tokens_per_second": 147217.71146206418, "data_wait_s": 8.276000244222814e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 318000, "loss": 0.5619775112718344, "lr": 0.0003, "grad_norm": 0.16456320219190762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46546627236, "tokens_per_second": 150476.19854476769, "data_wait_s": 1.5193999843177153e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 318100, "loss": 0.6014509499073029, "lr": 0.0003, "grad_norm": 0.15277714320537264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46561248304, "tokens_per_second": 146473.4475499082, "data_wait_s": 1.3373999991017627e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 318200, "loss": 0.5732059292495251, "lr": 0.0003, "grad_norm": 0.1512902765342098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46575750128, "tokens_per_second": 147097.79654431267, "data_wait_s": 1.1129000085929874e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a7845068e8486c97115368d41f03db2876c2dce37e2d5f0c67f3920a9fc741c
-size 65823
+oid sha256:fee097d3fa3dc378a1b0ba62ba64c161e767c9677d29a06df818925d7395a3f1
+size 67941