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": 75839.54600321392,
-      "required_pages_per_second_upper_estimate": 0.00657357963398654
+      "required_token_positions_per_second": 75969.7940652313,
+      "required_pages_per_second_upper_estimate": 0.006584869205891511
     },
     {
       "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": 9479.94325040174,
-      "required_pages_per_second_upper_estimate": 0.003781691100367696
+      "required_token_positions_per_second": 9496.224258153912,
+      "required_pages_per_second_upper_estimate": 0.0037881858377827957
     },
     {
       "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": 18959.88650080348,
-      "required_pages_per_second_upper_estimate": 0.004386036059355147
+      "required_token_positions_per_second": 18992.448516307824,
+      "required_pages_per_second_upper_estimate": 0.004393568708570177
     },
     {
       "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": 28439.82975120522,
-      "required_pages_per_second_upper_estimate": 0.001731836419593424
+      "required_token_positions_per_second": 28488.67277446174,
+      "required_pages_per_second_upper_estimate": 0.0017348107034502076
     },
     {
       "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": 28439.82975120522,
-      "required_pages_per_second_upper_estimate": 0.0021893768207352487
+      "required_token_positions_per_second": 28488.67277446174,
+      "required_pages_per_second_upper_estimate": 0.0021931368918716772
     },
     {
       "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": 28439.82975120522,
-      "required_pages_per_second_upper_estimate": 0.3531275035226693
+      "required_token_positions_per_second": 28488.67277446174,
+      "required_pages_per_second_upper_estimate": 0.353733970404432
     }
   ],
   "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": 0.12315990569510404,
+    "producer_batch_seconds": 0.10268041665648298,
     "producer_p95_seconds": 0.4178811950041563,
-    "gpu_update_seconds": 0.9721577183080126,
-    "producer_batches_per_second": 8.119525541661346,
-    "gpu_batches_per_second": 1.0286396756078278,
-    "producer_headroom": 7.893459424325124,
+    "gpu_update_seconds": 0.9704909814115543,
+    "producer_batches_per_second": 9.73895541683958,
+    "gpu_batches_per_second": 1.0304062780114922,
+    "producer_headroom": 9.451568400412018,
     "sustainable": true,
-    "buffer_seconds": 60.27377853509678,
+    "buffer_seconds": 60.17044084751637,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322764.41944456607,
-    "upload_bytes_per_second": 3641902.377202831,
+    "generation_bytes_per_second": 323802.1967723479,
+    "upload_bytes_per_second": 3634423.476470282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.25773084166275,
+    "estimated_full_shards_per_hour": 69.48041369798496,
     "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": 0.7187483379966579,
-      "bytes": 1132933,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.4302320659990073,
       "bytes": 11751486,
@@ -562,6 +556,12 @@
       "bytes": 11328528,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.9737502849966404,
+      "bytes": 1202420,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 40800, "loss": 0.6299373246729374, "lr": 0.0003, "grad_norm": 0.127283173593747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5970591672, "tokens_per_second": 150137.56871674163, "data_wait_s": 1.2154996511526406e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40900, "loss": 0.656970776617527, "lr": 0.0003, "grad_norm": 0.14349440390525414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5985111871, "tokens_per_second": 129470.72505672603, "data_wait_s": 1.4625002222601324e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41000, "loss": 0.633791621774435, "lr": 0.0003, "grad_norm": 0.13616520934006598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5999586229, "tokens_per_second": 148245.87926764408, "data_wait_s": 9.536001016385853e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 41100, "loss": 0.6907564885914326, "lr": 0.0003, "grad_norm": 0.14187858018878158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6014175298, "tokens_per_second": 148623.47853386478, "data_wait_s": 8.615003025624901e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1347835d5300c260d66bf2d72778bb22a7cee8a2288940341b6b20bc38f00305
-size 869035
+oid sha256:77f7c0edcfb205cf6a78dffb5e2a3527cea0fd73906a758c47d9413aa752c386
+size 871153