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": 75690.547853983,
-      "required_pages_per_second_upper_estimate": 0.00656066485204356
+      "required_token_positions_per_second": 75770.27840571137,
+      "required_pages_per_second_upper_estimate": 0.0065675756941921395
     },
     {
       "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": 9461.318481747874,
-      "required_pages_per_second_upper_estimate": 0.003774261401686562
+      "required_token_positions_per_second": 9471.284800713922,
+      "required_pages_per_second_upper_estimate": 0.0037782371153318663
     },
     {
       "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": 18922.63696349575,
-      "required_pages_per_second_upper_estimate": 0.004377419034468469
+      "required_token_positions_per_second": 18942.569601427844,
+      "required_pages_per_second_upper_estimate": 0.004382030099451614
     },
     {
       "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": 28383.955445243624,
-      "required_pages_per_second_upper_estimate": 0.0017284339675101904
+      "required_token_positions_per_second": 28413.854402141762,
+      "required_pages_per_second_upper_estimate": 0.0017302546571175835
     },
     {
       "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": 28383.955445243624,
-      "required_pages_per_second_upper_estimate": 0.0021850754619923466
+      "required_token_positions_per_second": 28413.854402141762,
+      "required_pages_per_second_upper_estimate": 0.0021873771664599755
     },
     {
       "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": 28383.955445243624,
-      "required_pages_per_second_upper_estimate": 0.35243373164189906
+      "required_token_positions_per_second": 28413.854402141762,
+      "required_pages_per_second_upper_estimate": 0.3528049766212022
     }
   ],
   "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": 0.13006275634381836,
-    "producer_p95_seconds": 0.3386017880002328,
-    "gpu_update_seconds": 0.9740714275477433,
-    "producer_batches_per_second": 7.688596090924903,
-    "gpu_batches_per_second": 1.026618758870212,
-    "producer_headroom": 7.489241770125219,
+    "producer_batch_seconds": 0.07690463121861058,
+    "producer_p95_seconds": 0.2982471040013479,
+    "gpu_update_seconds": 0.9730464444808292,
+    "producer_batches_per_second": 13.003118071750203,
+    "gpu_batches_per_second": 1.0277001736885765,
+    "producer_headroom": 12.652637806880952,
     "sustainable": true,
-    "buffer_seconds": 60.39242850796009,
+    "buffer_seconds": 60.32887955781141,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296320.8234938329,
+    "generation_bytes_per_second": 299825.36351798946,
     "upload_bytes_per_second": 5108532.204206022,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.58355072604409,
+    "estimated_full_shards_per_hour": 64.33554343371165,
     "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": [

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

@@ -236,3 +236,4 @@
 {"step": 260700, "loss": 0.6145157720893621, "lr": 0.0003, "grad_norm": 0.15188602167188195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38159524794, "tokens_per_second": 148843.88656316936, "data_wait_s": 9.1439978859853e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 260800, "loss": 0.5926761105656624, "lr": 0.0003, "grad_norm": 0.15779540370578413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38174072462, "tokens_per_second": 148441.49215839626, "data_wait_s": 1.4753997675143182e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 260900, "loss": 0.5752182491123676, "lr": 0.0003, "grad_norm": 0.13541259403785755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38188701425, "tokens_per_second": 149152.720374841, "data_wait_s": 1.499399877502583e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 261000, "loss": 0.5985351726412773, "lr": 0.0003, "grad_norm": 0.14187585540698935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38203433879, "tokens_per_second": 149347.4343050983, "data_wait_s": 1.337999856332317e-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:c75bbaac29358b3d65c383a70a1c5d2b97a26e009563255bc7ec844eab6e6d97
-size 508164
+oid sha256:c0999b83130e3735a3267fca0d5adeab7c65c4ff884cf8c3a13630b1420b6e5e
+size 510282