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": 75025.9953334272,
-      "required_pages_per_second_upper_estimate": 0.0065030631238547475
+      "required_token_positions_per_second": 75135.3888082297,
+      "required_pages_per_second_upper_estimate": 0.006512545072995402
     },
     {
       "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": 9378.2494166784,
-      "required_pages_per_second_upper_estimate": 0.003741123909637147
+      "required_token_positions_per_second": 9391.923601028713,
+      "required_pages_per_second_upper_estimate": 0.003746578746221762
     },
     {
       "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": 18756.4988333568,
-      "required_pages_per_second_upper_estimate": 0.00433898590199205
+      "required_token_positions_per_second": 18783.847202057426,
+      "required_pages_per_second_upper_estimate": 0.004345312465776089
     },
     {
       "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": 28134.748250035198,
-      "required_pages_per_second_upper_estimate": 0.0017132585568111025
+      "required_token_positions_per_second": 28175.77080308614,
+      "required_pages_per_second_upper_estimate": 0.0017157566150632025
     },
     {
       "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": 28134.748250035198,
-      "required_pages_per_second_upper_estimate": 0.0021658908022555334
+      "required_token_positions_per_second": 28175.77080308614,
+      "required_pages_per_second_upper_estimate": 0.0021690488319470896
     },
     {
       "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": 28134.748250035198,
-      "required_pages_per_second_upper_estimate": 0.34933941232024035
+      "required_token_positions_per_second": 28175.77080308614,
+      "required_pages_per_second_upper_estimate": 0.34984877513547985
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10366631672660276,
-    "producer_p95_seconds": 0.1993990360006137,
-    "gpu_update_seconds": 0.9826993920219425,
-    "producer_batches_per_second": 9.646334813238145,
-    "gpu_batches_per_second": 1.0176051884416666,
-    "producer_headroom": 9.479447356209223,
+    "producer_batch_seconds": 0.060204902648408165,
+    "producer_p95_seconds": 0.19866809799896146,
+    "gpu_update_seconds": 0.981268629462718,
+    "producer_batches_per_second": 16.609942978230865,
+    "gpu_batches_per_second": 1.0190889324032892,
+    "producer_headroom": 16.298815981702496,
     "sustainable": true,
-    "buffer_seconds": 60.92736230536043,
+    "buffer_seconds": 60.838655026688514,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313144.7536059806,
-    "upload_bytes_per_second": 5621954.086997323,
+    "generation_bytes_per_second": 313802.0088888452,
+    "upload_bytes_per_second": 5743684.237788523,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.19357448706211,
+    "estimated_full_shards_per_hour": 67.33460617064493,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 10983620,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 1.0362804550004512,
+      "bytes": 7507968,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 91100, "loss": 0.5345545094460249, "lr": 0.0003, "grad_norm": 0.21596472364712913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13331958244, "tokens_per_second": 149824.03475757226, "data_wait_s": 1.581699871167075e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91200, "loss": 0.6174646075814962, "lr": 0.0003, "grad_norm": 0.13663930383920614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13346623319, "tokens_per_second": 149101.99070978398, "data_wait_s": 1.4031000318937004e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91300, "loss": 0.661818029358983, "lr": 0.0003, "grad_norm": 0.13152827585548027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13361373514, "tokens_per_second": 150239.82422898128, "data_wait_s": 7.86600139690563e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 91400, "loss": 0.6057727262377739, "lr": 0.0003, "grad_norm": 0.12462032566448279, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13375999485, "tokens_per_second": 149018.34903248824, "data_wait_s": 1.842299934651237e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e5ccb90f2af337c17d6ac36b94c7ce7587508920eb946d0083adb2e63c9323d
-size 134382
+oid sha256:a6184b5875a12eab15edef93c92ea77e2f51a36fdcf726d4e357b203afe8c68d
+size 136500