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": 75630.60254063262,
-      "required_pages_per_second_upper_estimate": 0.006555468944212357
+      "required_token_positions_per_second": 75462.63472042159,
+      "required_pages_per_second_upper_estimate": 0.0065409099192670125
     },
     {
       "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": 9453.825317579078,
-      "required_pages_per_second_upper_estimate": 0.003771272266466841
+      "required_token_positions_per_second": 9432.829340052698,
+      "required_pages_per_second_upper_estimate": 0.00376289665711373
     },
     {
       "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": 18907.650635158156,
-      "required_pages_per_second_upper_estimate": 0.004373952211157938
+      "required_token_positions_per_second": 18865.658680105396,
+      "required_pages_per_second_upper_estimate": 0.0043642381113896496
     },
     {
       "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": 28361.475952737233,
-      "required_pages_per_second_upper_estimate": 0.0017270650843573335
+      "required_token_positions_per_second": 28298.488020158093,
+      "required_pages_per_second_upper_estimate": 0.0017232294497354058
     },
     {
       "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": 28361.475952737233,
-      "required_pages_per_second_upper_estimate": 0.002183344928431987
+      "required_token_positions_per_second": 28298.488020158093,
+      "required_pages_per_second_upper_estimate": 0.0021784959430202876
     },
     {
       "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": 28361.475952737233,
-      "required_pages_per_second_upper_estimate": 0.3521546115789914
+      "required_token_positions_per_second": 28298.488020158093,
+      "required_pages_per_second_upper_estimate": 0.35137251226340804
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0792378665312583,
-    "producer_p95_seconds": 0.2171078139999736,
-    "gpu_update_seconds": 0.9748434829722473,
-    "producer_batches_per_second": 12.620228733764721,
-    "gpu_batches_per_second": 1.0258056985220354,
-    "producer_headroom": 12.302747734729635,
+    "producer_batch_seconds": 0.16695900194537217,
+    "producer_p95_seconds": 0.2752427550003631,
+    "gpu_update_seconds": 0.9770133294861999,
+    "producer_batches_per_second": 5.989494356987071,
+    "gpu_batches_per_second": 1.0235274891550237,
+    "producer_headroom": 5.851815823658744,
     "sustainable": true,
-    "buffer_seconds": 60.440295944279335,
+    "buffer_seconds": 60.57482642814439,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 523269.04561499885,
-    "upload_bytes_per_second": 4017522.9884006456,
+    "generation_bytes_per_second": 498846.6434101604,
+    "upload_bytes_per_second": 4059163.4065738246,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.28135610902285,
+    "estimated_full_shards_per_hour": 107.04087712029084,
     "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": 1.6009747379994224,
-      "bytes": 7396060,
-      "pages": 37,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9060076790001403,
-      "bytes": 1026778,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.30779888699999,
       "bytes": 10925777,
@@ -562,6 +550,18 @@
       "bytes": 1160564,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.3003170710017002,
+      "bytes": 11354768,
+      "pages": 32,
+      "shards": 4
+    },
+    {
+      "seconds": 0.953995298999871,
+      "bytes": 345798,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 329800, "loss": 0.5932347755879164, "lr": 0.0003, "grad_norm": 0.15959028214622506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48274110536, "tokens_per_second": 148405.0384085669, "data_wait_s": 1.3793000107398257e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329900, "loss": 0.5818005800247192, "lr": 0.0003, "grad_norm": 0.16854894922756233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48288740098, "tokens_per_second": 148915.43229986343, "data_wait_s": 1.1409001672291197e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330000, "loss": 0.5456385742872953, "lr": 0.0003, "grad_norm": 0.14405760445189703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48303357762, "tokens_per_second": 148276.12502650137, "data_wait_s": 1.3793998732580803e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 330100, "loss": 0.6080137956887484, "lr": 0.0003, "grad_norm": 0.17845473445853496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48317901815, "tokens_per_second": 134522.147590586, "data_wait_s": 1.3312999726622365e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bbdc7631339fb8018df59a8d2542349ba28a9fc3096b7a8dbf2ebf0b0d83feb
-size 320214
+oid sha256:04faaf23b577765b989504bc0ccb92ccf42c28b6303fb1e0949450cf9db5c50e
+size 322332