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": 75116.97839899872,
-      "required_pages_per_second_upper_estimate": 0.006510949305384016
+      "required_token_positions_per_second": 75401.73314238236,
+      "required_pages_per_second_upper_estimate": 0.006535631124836211
     },
     {
       "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": 9389.62229987484,
-      "required_pages_per_second_upper_estimate": 0.0037456607227839634
+      "required_token_positions_per_second": 9425.216642797795,
+      "required_pages_per_second_upper_estimate": 0.003759859838358782
     },
     {
       "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": 18779.24459974968,
-      "required_pages_per_second_upper_estimate": 0.0043442477347352265
+      "required_token_positions_per_second": 18850.43328559559,
+      "required_pages_per_second_upper_estimate": 0.004360715984327598
     },
     {
       "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": 28168.866899624518,
-      "required_pages_per_second_upper_estimate": 0.001715336203564372
+      "required_token_positions_per_second": 28275.649928393388,
+      "required_pages_per_second_upper_estimate": 0.0017218387297691423
     },
     {
       "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": 28168.866899624518,
-      "required_pages_per_second_upper_estimate": 0.002168517350347386
+      "required_token_positions_per_second": 28275.649928393388,
+      "required_pages_per_second_upper_estimate": 0.0021767378034963565
     },
     {
       "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": 28168.866899624518,
-      "required_pages_per_second_upper_estimate": 0.34976305176036504
+      "required_token_positions_per_second": 28275.649928393388,
+      "required_pages_per_second_upper_estimate": 0.3510889395978667
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08428292574231477,
-    "producer_p95_seconds": 0.21030219300155295,
-    "gpu_update_seconds": 0.9815091284473547,
-    "producer_batches_per_second": 11.864799319584414,
-    "gpu_batches_per_second": 1.018839225246836,
-    "producer_headroom": 11.645408839368065,
+    "producer_batch_seconds": 0.07115390350013229,
+    "producer_p95_seconds": 0.19268069399913657,
+    "gpu_update_seconds": 0.9778024579458694,
+    "producer_batches_per_second": 14.054042727229165,
+    "gpu_batches_per_second": 1.0227014586369136,
+    "producer_headroom": 13.742077522760948,
     "sustainable": true,
-    "buffer_seconds": 60.853565963735996,
+    "buffer_seconds": 60.6237523926439,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268096.43718736334,
-    "upload_bytes_per_second": 3661950.9345863685,
+    "generation_bytes_per_second": 268058.95697252767,
+    "upload_bytes_per_second": 3666752.2604912724,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.52725445476222,
+    "estimated_full_shards_per_hour": 57.519212073153234,
     "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": 1.0704197729999123,
-      "bytes": 6206676,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 2.8963105739999264,
       "bytes": 7237253,
@@ -562,6 +556,12 @@
       "bytes": 12063808,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.8872203939990868,
+      "bytes": 6084404,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -87,3 +87,4 @@
 {"step": 245800, "loss": 0.6115876995027065, "lr": 0.0003, "grad_norm": 0.16180542161259748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35977951781, "tokens_per_second": 148257.8724533782, "data_wait_s": 1.6004998542484827e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 245900, "loss": 0.61958840303123, "lr": 0.0003, "grad_norm": 0.16966736014105382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35992654748, "tokens_per_second": 121703.00145173802, "data_wait_s": 1.2460999641916715e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246000, "loss": 0.6332357935607433, "lr": 0.0003, "grad_norm": 0.15884492809949485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36007415822, "tokens_per_second": 151095.6949204287, "data_wait_s": 1.1965999874519184e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 246100, "loss": 0.5597578007727861, "lr": 0.0003, "grad_norm": 0.13582123623624906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36022129649, "tokens_per_second": 149542.31262298633, "data_wait_s": 1.5158000678638928e-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:fc1f1223f89f8b4967206dcc90a303587a161527e3c19049c348ebdb8e7aec9f
-size 190233
+oid sha256:1074608dbdc7deed1541c1cae404b544522570ba41eaad42c6e06a00cf444d56
+size 192351