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": 75329.26328360828,
-      "required_pages_per_second_upper_estimate": 0.0065293496211508515
+      "required_token_positions_per_second": 76244.84033504393,
+      "required_pages_per_second_upper_estimate": 0.006608709519460472
     },
     {
       "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": 9416.157910451035,
-      "required_pages_per_second_upper_estimate": 0.0037562461745855413
+      "required_token_positions_per_second": 9530.605041880492,
+      "required_pages_per_second_upper_estimate": 0.003801900846449853
     },
     {
       "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": 18832.31582090207,
-      "required_pages_per_second_upper_estimate": 0.00435652482772724
+      "required_token_positions_per_second": 19061.210083760983,
+      "required_pages_per_second_upper_estimate": 0.0044094754870382105
     },
     {
       "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": 28248.473731353104,
-      "required_pages_per_second_upper_estimate": 0.0017201838419518755
+      "required_token_positions_per_second": 28591.815125641475,
+      "required_pages_per_second_upper_estimate": 0.001741091531490959
     },
     {
       "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": 28248.473731353104,
-      "required_pages_per_second_upper_estimate": 0.002174645704619667
+      "required_token_positions_per_second": 28591.815125641475,
+      "required_pages_per_second_upper_estimate": 0.0022010770755817952
     },
     {
       "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": 28248.473731353104,
-      "required_pages_per_second_upper_estimate": 0.35075150218350704
+      "required_token_positions_per_second": 28591.815125641475,
+      "required_pages_per_second_upper_estimate": 0.3550146532108407
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09384961011713244,
+    "producer_batch_seconds": 0.14631836382042707,
     "producer_p95_seconds": 0.384045917999174,
-    "gpu_update_seconds": 0.9787431442468825,
-    "producer_batches_per_second": 10.655345277960274,
-    "gpu_batches_per_second": 1.0217185232694264,
-    "producer_headroom": 10.42884614038701,
+    "gpu_update_seconds": 0.9669900241906975,
+    "producer_batches_per_second": 6.834412126336209,
+    "gpu_batches_per_second": 1.0341368318012687,
+    "producer_headroom": 6.608808347375047,
     "sustainable": true,
-    "buffer_seconds": 60.682074943306716,
+    "buffer_seconds": 60.92037152401394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481789.2747620921,
-    "upload_bytes_per_second": 5294670.331594702,
+    "generation_bytes_per_second": 479315.0567932494,
+    "upload_bytes_per_second": 5251794.077648563,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.38076288363526,
+    "estimated_full_shards_per_hour": 102.84985330436812,
     "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": 1.1408973270008573,
-      "bytes": 5952533,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.4523514239990618,
       "bytes": 13500548,
@@ -562,6 +556,12 @@
       "bytes": 16078869,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.8470632479984488,
+      "bytes": 947755,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -321,3 +321,4 @@
 {"step": 147100, "loss": 0.5724699255079031, "lr": 0.0003, "grad_norm": 0.1468198003576462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21527672095, "tokens_per_second": 147824.88346666755, "data_wait_s": 1.1464999261079356e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147200, "loss": 0.6118672527372837, "lr": 0.0003, "grad_norm": 0.1628369852678089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21542244994, "tokens_per_second": 149631.85286732164, "data_wait_s": 9.551000403007492e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 147300, "loss": 0.6135005541145802, "lr": 0.0003, "grad_norm": 0.14085108673209615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21556893328, "tokens_per_second": 150064.4933955498, "data_wait_s": 1.6808997315820307e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 147400, "loss": 0.577962338924408, "lr": 0.0003, "grad_norm": 0.14103880552126036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21571622651, "tokens_per_second": 150578.31751383236, "data_wait_s": 1.5854002413107082e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1978914881f02bfaa5e9f9b00cd52fe8e6a7efe59f8cdaf1386cfb1b5fe1e6aa
-size 688977
+oid sha256:52be32e74a278742995661ec83c1bf7641a9a6eec94cfa34f6104fd7e4d1402a
+size 691095