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": 75504.56331335989,
-      "required_pages_per_second_upper_estimate": 0.006544544183435857
+      "required_token_positions_per_second": 75636.41396792672,
+      "required_pages_per_second_upper_estimate": 0.006555972664001282
     },
     {
       "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": 9438.070414169986,
-      "required_pages_per_second_upper_estimate": 0.0037649873999401574
+      "required_token_positions_per_second": 9454.55174599084,
+      "required_pages_per_second_upper_estimate": 0.003771562049621366
     },
     {
       "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": 18876.140828339972,
-      "required_pages_per_second_upper_estimate": 0.004366662971904211
+      "required_token_positions_per_second": 18909.10349198168,
+      "required_pages_per_second_upper_estimate": 0.004374288304014654
     },
     {
       "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": 28314.21124250996,
-      "required_pages_per_second_upper_estimate": 0.001724186911483262
+      "required_token_positions_per_second": 28363.65523797252,
+      "required_pages_per_second_upper_estimate": 0.0017271977913943365
     },
     {
       "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": 28314.21124250996,
-      "required_pages_per_second_upper_estimate": 0.0021797063602015995
+      "required_token_positions_per_second": 28363.65523797252,
+      "required_pages_per_second_upper_estimate": 0.002183512695842047
     },
     {
       "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": 28314.21124250996,
-      "required_pages_per_second_upper_estimate": 0.3515677420627781
+      "required_token_positions_per_second": 28363.65523797252,
+      "required_pages_per_second_upper_estimate": 0.35218167100801523
     }
   ],
   "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": 0.11990805596097687,
-    "producer_p95_seconds": 0.2960198490000039,
-    "gpu_update_seconds": 0.9764707822229661,
-    "producer_batches_per_second": 8.3397232319857,
-    "gpu_batches_per_second": 1.0240961820930974,
-    "producer_headroom": 8.143496067860118,
+    "producer_batch_seconds": 0.08550059055477277,
+    "producer_p95_seconds": 0.22137790600027074,
+    "gpu_update_seconds": 0.9747685821179204,
+    "producer_batches_per_second": 11.695825648822709,
+    "gpu_batches_per_second": 1.0258845210493532,
+    "producer_headroom": 11.400723384401319,
     "sustainable": true,
-    "buffer_seconds": 60.5411884978239,
+    "buffer_seconds": 60.43565209131106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366851.73369749496,
-    "upload_bytes_per_second": 4137697.6116614225,
+    "generation_bytes_per_second": 434795.3200518696,
+    "upload_bytes_per_second": 4198238.345037328,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.71784218019138,
+    "estimated_full_shards_per_hour": 93.29695416609827,
     "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": 1.2160326559996975,
-      "bytes": 2369678,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.405440059999819,
       "bytes": 15806866,
@@ -562,6 +556,12 @@
       "bytes": 15458880,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.9607973950005544,
+      "bytes": 7512102,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -146,3 +146,4 @@
 {"step": 329600, "loss": 0.6107210144400597, "lr": 0.0003, "grad_norm": 0.144386048113135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48245021257, "tokens_per_second": 146190.4000532596, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329700, "loss": 0.5373153388500214, "lr": 0.0003, "grad_norm": 0.15336776926794268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48259548074, "tokens_per_second": 147798.01532501896, "data_wait_s": 1.5064999388414435e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:10c2680feb639e10635fc1c9594591ba5abf3cd8640fff11fc2daa766eaed3cb
-size 315195
+oid sha256:50ebf408f3e01bb70026b30244a6fe1b888c910aca6561e776da101a4192d71a
+size 317313