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": 73956.98965365934,
-      "required_pages_per_second_upper_estimate": 0.0064104044209025955
+      "required_token_positions_per_second": 75600.19049404793,
+      "required_pages_per_second_upper_estimate": 0.006552832905093023
     },
     {
       "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": 9244.623706707418,
-      "required_pages_per_second_upper_estimate": 0.0036878186160473185
+      "required_token_positions_per_second": 9450.023811755991,
+      "required_pages_per_second_upper_estimate": 0.003769755788956435
     },
     {
       "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": 18489.247413414836,
-      "required_pages_per_second_upper_estimate": 0.004277161989452824
+      "required_token_positions_per_second": 18900.047623511982,
+      "required_pages_per_second_upper_estimate": 0.004372193388222031
     },
     {
       "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": 27733.871120122258,
-      "required_pages_per_second_upper_estimate": 0.0016888472428391581
+      "required_token_positions_per_second": 28350.07143526797,
+      "required_pages_per_second_upper_estimate": 0.0017263706090783343
     },
     {
       "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": 27733.871120122258,
-      "required_pages_per_second_upper_estimate": 0.0021350301710959156
+      "required_token_positions_per_second": 28350.07143526797,
+      "required_pages_per_second_upper_estimate": 0.002182466977107478
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27733.871120122258,
-      "required_pages_per_second_upper_estimate": 0.3443618600161697
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28350.07143526797,
+      "required_pages_per_second_upper_estimate": 0.3520130056404879
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3238106990629035,
-    "producer_p95_seconds": 0.4167272660051822,
-    "gpu_update_seconds": 0.996903745613069,
-    "producer_batches_per_second": 3.0882240855350487,
-    "gpu_batches_per_second": 1.003105870953496,
-    "producer_headroom": 3.078662158162385,
+    "producer_batch_seconds": 0.1922335974293219,
+    "producer_p95_seconds": 0.4000828360003652,
+    "gpu_update_seconds": 0.9752356378758685,
+    "producer_batches_per_second": 5.202004297753767,
+    "gpu_batches_per_second": 1.0253932087408844,
+    "producer_headroom": 5.073179979552905,
     "sustainable": true,
-    "buffer_seconds": 60.81112848239721,
+    "buffer_seconds": 60.46460954830385,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360659.31272286805,
-    "upload_bytes_per_second": 3807971.006932565,
+    "generation_bytes_per_second": 372329.10340258345,
+    "upload_bytes_per_second": 3810591.506888325,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.38909279121906,
+    "estimated_full_shards_per_hour": 79.89315821226242,
     "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": 1.0204439990011451,
-      "bytes": 7601581,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.844439466996846,
       "bytes": 1090867,
@@ -562,6 +556,12 @@
       "bytes": 10363400,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1336681269967812,
+      "bytes": 8324153,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -357,3 +357,4 @@
 {"step": 35700, "loss": 0.6167456582188606, "lr": 0.0003, "grad_norm": 0.1455468630375116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5223942532, "tokens_per_second": 149018.186610883, "data_wait_s": 1.1085998266935349e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35800, "loss": 0.6350256651639938, "lr": 0.0003, "grad_norm": 0.12678470938493638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5238589021, "tokens_per_second": 148927.44061422633, "data_wait_s": 1.4830999134574085e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35900, "loss": 0.6602982357144356, "lr": 0.0003, "grad_norm": 0.12721359935238322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5253246401, "tokens_per_second": 148978.6253565894, "data_wait_s": 1.0269999620504677e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 36000, "loss": 0.6540908552706242, "lr": 0.0003, "grad_norm": 0.1320960322630182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5267859951, "tokens_per_second": 149059.11328397377, "data_wait_s": 1.0726005712058395e-05, "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:f212360ea48a37df3387285563b98194faa53e3c3d57b6e1367b744070929cb4
-size 760234
+oid sha256:1971cbef22926b8580326129703cf68695a389475e356338d77584d900aa4d5f
+size 762352