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": 75933.99341993597,
-      "required_pages_per_second_upper_estimate": 0.006581766096693218
+      "required_token_positions_per_second": 75708.09017882668,
+      "required_pages_per_second_upper_estimate": 0.006562185376300394
     },
     {
       "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": 9491.749177491996,
-      "required_pages_per_second_upper_estimate": 0.0037864006611983392
+      "required_token_positions_per_second": 9463.511272353335,
+      "required_pages_per_second_upper_estimate": 0.0037751361386442217
     },
     {
       "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": 18983.498354983993,
-      "required_pages_per_second_upper_estimate": 0.004391498246265368
+      "required_token_positions_per_second": 18927.02254470667,
+      "required_pages_per_second_upper_estimate": 0.004378433561498023
     },
     {
       "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": 28475.24753247599,
-      "required_pages_per_second_upper_estimate": 0.0017339931766500768
+      "required_token_positions_per_second": 28390.533817060004,
+      "required_pages_per_second_upper_estimate": 0.0017288345558397584
     },
     {
       "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": 28475.24753247599,
-      "required_pages_per_second_upper_estimate": 0.0021921033795801663
+      "required_token_positions_per_second": 28390.533817060004,
+      "required_pages_per_second_upper_estimate": 0.0021855818832648703
     },
     {
       "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": 28475.24753247599,
-      "required_pages_per_second_upper_estimate": 0.35356727383036357
+      "required_token_positions_per_second": 28390.533817060004,
+      "required_pages_per_second_upper_estimate": 0.3525154130034643
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07296771318753059,
-    "producer_p95_seconds": 0.17815108800004964,
-    "gpu_update_seconds": 0.9709485393750303,
-    "producer_batches_per_second": 13.704691517876558,
-    "gpu_batches_per_second": 1.0299207006827253,
-    "producer_headroom": 13.306550211867611,
+    "producer_batch_seconds": 0.31357948278120773,
+    "producer_p95_seconds": 0.18834892100039724,
+    "gpu_update_seconds": 0.9738457254152152,
+    "producer_batches_per_second": 3.188984148869603,
+    "gpu_batches_per_second": 1.026856691878617,
+    "producer_headroom": 3.105578581793541,
     "sustainable": true,
-    "buffer_seconds": 60.19880944125188,
+    "buffer_seconds": 60.37843497574334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 468602.7926639869,
-    "upload_bytes_per_second": 4826734.250837589,
+    "generation_bytes_per_second": 422421.0882053992,
+    "upload_bytes_per_second": 4714265.580201424,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.55125079097466,
+    "estimated_full_shards_per_hour": 90.64173206922038,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 7410356,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.306584923000628,
+      "bytes": 936913,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -42,3 +42,4 @@
 {"step": 241300, "loss": 0.6611749287694693, "lr": 0.0003, "grad_norm": 0.15601242358858483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35319001314, "tokens_per_second": 150003.28089799767, "data_wait_s": 1.2620000234164763e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241400, "loss": 0.6207055300474167, "lr": 0.0003, "grad_norm": 0.13743057772118955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35333684821, "tokens_per_second": 149884.36847653115, "data_wait_s": 1.9915999473596457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241500, "loss": 0.58491569198668, "lr": 0.0003, "grad_norm": 0.15787964599663515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35348409349, "tokens_per_second": 149601.08783378193, "data_wait_s": 1.3154000043869019e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 241600, "loss": 0.6268334165215492, "lr": 0.0003, "grad_norm": 0.14220717144270645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35363108475, "tokens_per_second": 148962.41087267562, "data_wait_s": 9.933000001183245e-06, "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 sha256:455ca05f500dcade40ec102ba75fb58f81dde79b8579e38578c60e4f29e20900
-size 94140
+oid sha256:af8cc4a393f39057b5ab1e63187cc90556bce713642d0bbeccb77215a995f628
+size 96258