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": 75351.31612105282,
-      "required_pages_per_second_upper_estimate": 0.006531261105208138
+      "required_token_positions_per_second": 75425.56319680403,
+      "required_pages_per_second_upper_estimate": 0.00653769665355676
     },
     {
       "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": 9418.914515131602,
-      "required_pages_per_second_upper_estimate": 0.0037573458254075323
+      "required_token_positions_per_second": 9428.195399600503,
+      "required_pages_per_second_upper_estimate": 0.0037610481089837655
     },
     {
       "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": 18837.829030263205,
-      "required_pages_per_second_upper_estimate": 0.004357800211683768
+      "required_token_positions_per_second": 18856.390799201006,
+      "required_pages_per_second_upper_estimate": 0.004362094150251552
     },
     {
       "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": 28256.743545394806,
-      "required_pages_per_second_upper_estimate": 0.0017206874302386542
+      "required_token_positions_per_second": 28284.58619880151,
+      "required_pages_per_second_upper_estimate": 0.0017223829017520096
     },
     {
       "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": 28256.743545394806,
-      "required_pages_per_second_upper_estimate": 0.0021752823377969074
+      "required_token_positions_per_second": 28284.58619880151,
+      "required_pages_per_second_upper_estimate": 0.002177425742329553
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28256.743545394806,
-      "required_pages_per_second_upper_estimate": 0.3508541855966178
+      "required_token_positions_per_second": 28284.58619880151,
+      "required_pages_per_second_upper_estimate": 0.3511998981685624
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.49937937652339315,
-    "producer_p95_seconds": 0.24167164899881755,
-    "gpu_update_seconds": 0.9784566985074961,
-    "producer_batches_per_second": 2.0024855791239418,
-    "gpu_batches_per_second": 1.0220176340203562,
-    "producer_headroom": 1.9593454285584835,
+    "producer_batch_seconds": 0.08177704960159815,
+    "producer_p95_seconds": 0.21488423800110468,
+    "gpu_update_seconds": 0.9774935297151887,
+    "producer_batches_per_second": 12.228369754983888,
+    "gpu_batches_per_second": 1.0230246744358185,
+    "producer_headroom": 11.953152314461658,
     "sustainable": true,
-    "buffer_seconds": 60.664315307464754,
+    "buffer_seconds": 60.604598842341694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366770.8353441031,
-    "upload_bytes_per_second": 5231635.719438456,
+    "generation_bytes_per_second": 365186.2136648354,
+    "upload_bytes_per_second": 5354591.978705638,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.70048327677078,
+    "estimated_full_shards_per_hour": 78.36046035250469,
     "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": 0.9598245239999414,
-      "bytes": 1062466,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1868087180000657,
       "bytes": 11845509,
@@ -562,6 +556,12 @@
       "bytes": 11353142,
       "pages": 26,
       "shards": 4
+    },
+    {
+      "seconds": 1.1774445949995425,
+      "bytes": 12065836,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 165800, "loss": 0.6403736751526594, "lr": 0.0003, "grad_norm": 0.14658243310423927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24265691388, "tokens_per_second": 148917.27265474034, "data_wait_s": 9.196999599225819e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 165900, "loss": 0.5704298634082079, "lr": 0.0003, "grad_norm": 0.13297813866340955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24280196012, "tokens_per_second": 147638.1694864609, "data_wait_s": 1.1963000360992737e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 166000, "loss": 0.5583077687770128, "lr": 0.0003, "grad_norm": 0.14372924377867893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24294726970, "tokens_per_second": 148583.72982506774, "data_wait_s": 1.3250000847619958e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 166100, "loss": 0.638989269733429, "lr": 0.0003, "grad_norm": 0.14559203971628143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24309405732, "tokens_per_second": 150312.51889479955, "data_wait_s": 1.1656000424409285e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:813295d908336e5348a70156e5039979620d6553a39e050116b15dfdda1adcbc
-size 234711
+oid sha256:c10c057e68ae5e07ed2923a92478fdc70f424cc40a2c3bbff54e82d2e5de2123
+size 236829