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": 74422.98495920842,
-      "required_pages_per_second_upper_estimate": 0.006450795712933285
+      "required_token_positions_per_second": 74695.80482014333,
+      "required_pages_per_second_upper_estimate": 0.0064744430470235065
     },
     {
       "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": 9302.873119901053,
-      "required_pages_per_second_upper_estimate": 0.0037110551778765967
+      "required_token_positions_per_second": 9336.975602517916,
+      "required_pages_per_second_upper_estimate": 0.0037246591680700158
     },
     {
       "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": 18605.746239802105,
-      "required_pages_per_second_upper_estimate": 0.004304111942628201
+      "required_token_positions_per_second": 18673.95120503583,
+      "required_pages_per_second_upper_estimate": 0.004319889960968633
     },
     {
       "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": 27908.619359703156,
-      "required_pages_per_second_upper_estimate": 0.0016994884937964782
+      "required_token_positions_per_second": 28010.92680755375,
+      "required_pages_per_second_upper_estimate": 0.0017057184806048842
     },
     {
       "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": 27908.619359703156,
-      "required_pages_per_second_upper_estimate": 0.0021484827743129408
+      "required_token_positions_per_second": 28010.92680755375,
+      "required_pages_per_second_upper_estimate": 0.002156358684853621
     },
     {
       "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": 27908.619359703156,
-      "required_pages_per_second_upper_estimate": 0.34653164830702854
+      "required_token_positions_per_second": 28010.92680755375,
+      "required_pages_per_second_upper_estimate": 0.34780196440833094
     }
   ],
   "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": 0.10948076403909113,
-    "producer_p95_seconds": 0.22668656700261636,
-    "gpu_update_seconds": 0.9906616892672426,
-    "producer_batches_per_second": 9.134024673438894,
-    "gpu_batches_per_second": 1.009426336794819,
-    "producer_headroom": 9.04872831279765,
+    "producer_batch_seconds": 0.06814396046098636,
+    "producer_p95_seconds": 0.21162632399864378,
+    "gpu_update_seconds": 0.9870433845317864,
+    "producer_batches_per_second": 14.674814807286083,
+    "gpu_batches_per_second": 1.0131266929815448,
+    "producer_headroom": 14.48467887476083,
     "sustainable": true,
-    "buffer_seconds": 60.4303630453018,
+    "buffer_seconds": 60.209646456438975,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372943.28194573626,
+    "generation_bytes_per_second": 373238.44974129065,
     "upload_bytes_per_second": 4844899.91413792,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.02494663027827,
+    "estimated_full_shards_per_hour": 80.08828276804962,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -161,3 +161,4 @@
 {"step": 253200, "loss": 0.5419284403324127, "lr": 0.0003, "grad_norm": 0.15212501898844985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37061814354, "tokens_per_second": 149998.83716176712, "data_wait_s": 1.1742999049602076e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253300, "loss": 0.5990062914788723, "lr": 0.0003, "grad_norm": 0.1535534124407291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37076348893, "tokens_per_second": 141027.34075947746, "data_wait_s": 1.7050999304046854e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253400, "loss": 0.5942687429487705, "lr": 0.0003, "grad_norm": 0.13455753987219998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37090988499, "tokens_per_second": 149678.50081303433, "data_wait_s": 1.1964002624154091e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 253500, "loss": 0.5892300484701991, "lr": 0.0003, "grad_norm": 0.15466313150434347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37105693694, "tokens_per_second": 150293.5501557706, "data_wait_s": 1.0577001376077533e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:781ce2ba3fd849ceafd11c078faacb6ddc04ddcdcab58bfc2b6c67aec671eabf
-size 347748
+oid sha256:cfdd28f7e205971f18c4ed289c02f59f05bbfe68c447bee78e027813d6feb9bf
+size 349866