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": 77347.07578949745,
-      "required_pages_per_second_upper_estimate": 0.005351050173059398
+      "required_token_positions_per_second": 77155.68582227238,
+      "required_pages_per_second_upper_estimate": 0.005337809371040861
     },
     {
       "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": 9668.384473687182,
-      "required_pages_per_second_upper_estimate": 0.004047950680306331
+      "required_token_positions_per_second": 9644.460727784048,
+      "required_pages_per_second_upper_estimate": 0.004037934307481314
     },
     {
       "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": 19336.768947374363,
-      "required_pages_per_second_upper_estimate": 0.004027339026848629
+      "required_token_positions_per_second": 19288.921455568096,
+      "required_pages_per_second_upper_estimate": 0.004017373656128596
     },
     {
       "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": 29005.153421061546,
-      "required_pages_per_second_upper_estimate": 0.0017388146592877418
+      "required_token_positions_per_second": 28933.382183352143,
+      "required_pages_per_second_upper_estimate": 0.0017345120831753977
     },
     {
       "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": 29005.153421061546,
-      "required_pages_per_second_upper_estimate": 0.0022550230983007257
+      "required_token_positions_per_second": 28933.382183352143,
+      "required_pages_per_second_upper_estimate": 0.0022494432002571315
     },
     {
       "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": 29005.153421061546,
-      "required_pages_per_second_upper_estimate": 0.3720469647780499
+      "required_token_positions_per_second": 28933.382183352143,
+      "required_pages_per_second_upper_estimate": 0.3711263604026647
     }
   ],
   "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": 0.19388922930443186,
-    "producer_p95_seconds": 0.3142145050005638,
-    "gpu_update_seconds": 0.9532099209626633,
-    "producer_batches_per_second": 5.1575840679105855,
-    "gpu_batches_per_second": 1.0490868569539042,
-    "producer_headroom": 4.916260301731341,
+    "producer_batch_seconds": 0.3744751613436961,
+    "producer_p95_seconds": 0.30265087600128027,
+    "gpu_update_seconds": 0.9555744235082294,
+    "producer_batches_per_second": 2.6704040834425133,
+    "gpu_batches_per_second": 1.0464909643862899,
+    "producer_headroom": 2.5517698425696014,
     "sustainable": true,
-    "buffer_seconds": 60.05222502064779,
+    "buffer_seconds": 60.20118868101845,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427890.53004651173,
-    "upload_bytes_per_second": 5652003.746325318,
+    "generation_bytes_per_second": 409028.67553819035,
+    "upload_bytes_per_second": 5639558.632833212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.81534696623339,
+    "estimated_full_shards_per_hour": 87.76803207024844,
     "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": 1.2949794519990974,
-      "bytes": 7882340,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.516641152997181,
       "bytes": 1120266,
@@ -562,6 +556,12 @@
       "bytes": 6323168,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1956154220024473,
+      "bytes": 6398380,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 720400, "loss": 0.5309709310531616, "lr": 8e-05, "grad_norm": 0.2127561535188853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105446489981, "tokens_per_second": 153689.68463523037, "data_wait_s": 1.5589997929055244e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720500, "loss": 0.5453026294708252, "lr": 8e-05, "grad_norm": 0.2268957195603477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105461179941, "tokens_per_second": 153552.91413386018, "data_wait_s": 2.015000063693151e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720600, "loss": 0.5363257527351379, "lr": 8e-05, "grad_norm": 0.23791260345480333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105475828610, "tokens_per_second": 152334.94725441295, "data_wait_s": 1.4915000065229833e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720700, "loss": 0.5245054364204407, "lr": 8e-05, "grad_norm": 0.2286034159329082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105490324667, "tokens_per_second": 150841.09056894577, "data_wait_s": 1.5284000255633146e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:327533b8dca6a4452fa8ad407828aed2d341f1a64e51f69a2cee0e2e7cb346df
-size 759654
+oid sha256:bf77bbcee400cbb4b43a726003221864ed8247d22328fb3644c0ca4ad2f8c0d6
+size 761772