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": 77297.65395203981,
-      "required_pages_per_second_upper_estimate": 0.005347631055669616
+      "required_token_positions_per_second": 77205.33040770852,
+      "required_pages_per_second_upper_estimate": 0.005341243898652641
     },
     {
       "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": 9662.206744004976,
-      "required_pages_per_second_upper_estimate": 0.004045364193894057
+      "required_token_positions_per_second": 9650.666300963565,
+      "required_pages_per_second_upper_estimate": 0.0040405324513844735
     },
     {
       "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": 19324.413488009952,
-      "required_pages_per_second_upper_estimate": 0.004024765710498397
+      "required_token_positions_per_second": 19301.33260192713,
+      "required_pages_per_second_upper_estimate": 0.004019958570611244
     },
     {
       "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": 28986.620232014928,
-      "required_pages_per_second_upper_estimate": 0.0017377036229029382
+      "required_token_positions_per_second": 28951.998902890697,
+      "required_pages_per_second_upper_estimate": 0.0017356281270856492
     },
     {
       "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": 28986.620232014928,
-      "required_pages_per_second_upper_estimate": 0.0022535822243712348
+      "required_token_positions_per_second": 28951.998902890697,
+      "required_pages_per_second_upper_estimate": 0.0022508905683149586
     },
     {
       "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": 28986.620232014928,
-      "required_pages_per_second_upper_estimate": 0.3718092409283479
+      "required_token_positions_per_second": 28951.998902890697,
+      "required_pages_per_second_upper_estimate": 0.3713651556918292
     }
   ],
   "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": 0.09289110372660048,
-    "producer_p95_seconds": 0.25378325899873744,
-    "gpu_update_seconds": 0.953819375239323,
-    "producer_batches_per_second": 10.765293552149258,
-    "gpu_batches_per_second": 1.0484165303824844,
-    "producer_headroom": 10.268145570178918,
+    "producer_batch_seconds": 0.11462363989841151,
+    "producer_p95_seconds": 0.2611820820020512,
+    "gpu_update_seconds": 0.9549599698706641,
+    "producer_batches_per_second": 8.724203845614035,
+    "gpu_batches_per_second": 1.0471643121705259,
+    "producer_headroom": 8.33126544155311,
     "sustainable": true,
-    "buffer_seconds": 60.090620640077354,
+    "buffer_seconds": 60.16247810185184,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341224.0350326031,
-    "upload_bytes_per_second": 3915800.367332622,
+    "generation_bytes_per_second": 329119.0345114509,
+    "upload_bytes_per_second": 3863831.083724247,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.21873462899751,
+    "estimated_full_shards_per_hour": 70.62128330714842,
     "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": 1.3117666090001876,
-      "bytes": 13264605,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.0446096289997513,
       "bytes": 5930135,
@@ -562,6 +556,12 @@
       "bytes": 5692225,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0190912239995669,
+      "bytes": 6330148,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -279,3 +279,4 @@
 {"step": 552900, "loss": 0.5635361671447754, "lr": 0.0003, "grad_norm": 0.15189729263302537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80926877470, "tokens_per_second": 152746.83972867098, "data_wait_s": 1.4863999240333214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553000, "loss": 0.5958641767501831, "lr": 0.0003, "grad_norm": 0.16391493848202127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80941456564, "tokens_per_second": 151671.7997913953, "data_wait_s": 1.3291999493958429e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553100, "loss": 0.5691505670547485, "lr": 0.0003, "grad_norm": 0.15094018828179556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80956032059, "tokens_per_second": 152471.27029102138, "data_wait_s": 1.2281001545488834e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 553200, "loss": 0.5633944869041443, "lr": 0.0003, "grad_norm": 0.16626660762691406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80970770951, "tokens_per_second": 153594.97259423923, "data_wait_s": 1.8889997591031715e-05, "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 sha256:48652e18fb93d4cda063a9492619f358efd9ccc889bc53b69162485360c8e6ae
-size 599238
+oid sha256:6167f99311e52f00a12ae4080403b960ff95383cdd3828975f4c1a3ac6f73164
+size 601356