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": 75747.35037412556,
-      "required_pages_per_second_upper_estimate": 0.006565588350524846
+      "required_token_positions_per_second": 75580.90293211547,
+      "required_pages_per_second_upper_estimate": 0.006551161108108596
     },
     {
       "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": 9468.418796765694,
-      "required_pages_per_second_upper_estimate": 0.0037770938235063404
+      "required_token_positions_per_second": 9447.612866514433,
+      "required_pages_per_second_upper_estimate": 0.003768794026852734
     },
     {
       "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": 18936.83759353139,
-      "required_pages_per_second_upper_estimate": 0.004380704100304662
+      "required_token_positions_per_second": 18895.225733028867,
+      "required_pages_per_second_upper_estimate": 0.004371077928721136
     },
     {
       "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": 28405.256390297083,
-      "required_pages_per_second_upper_estimate": 0.0017297310833065244
+      "required_token_positions_per_second": 28342.8385995433,
+      "required_pages_per_second_upper_estimate": 0.001725930167330458
     },
     {
       "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": 28405.256390297083,
-      "required_pages_per_second_upper_estimate": 0.0021867152677073516
+      "required_token_positions_per_second": 28342.8385995433,
+      "required_pages_per_second_upper_estimate": 0.002181910173391622
     },
     {
       "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": 28405.256390297083,
-      "required_pages_per_second_upper_estimate": 0.35269821808978585
+      "required_token_positions_per_second": 28342.8385995433,
+      "required_pages_per_second_upper_estimate": 0.3519231980275314
     }
   ],
   "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": 0.12045250370340455,
-    "producer_p95_seconds": 0.39593479999894043,
-    "gpu_update_seconds": 0.9733409767582399,
-    "producer_batches_per_second": 8.302027515030685,
-    "gpu_batches_per_second": 1.0273891923573888,
-    "producer_headroom": 8.08070357055375,
+    "producer_batch_seconds": 0.14759950065609928,
+    "producer_p95_seconds": 0.4002699670018046,
+    "gpu_update_seconds": 0.9754845091784669,
+    "producer_batches_per_second": 6.775090671410593,
+    "gpu_batches_per_second": 1.0251316044395,
+    "producer_headroom": 6.608995998240572,
     "sustainable": true,
-    "buffer_seconds": 60.347140559010874,
+    "buffer_seconds": 60.48003956906494,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326761.38953539263,
-    "upload_bytes_per_second": 3248294.7884424794,
+    "generation_bytes_per_second": 347702.4335801336,
+    "upload_bytes_per_second": 3294452.632396012,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.11538757845244,
+    "estimated_full_shards_per_hour": 74.60884814789777,
     "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": 0.908005076002155,
-      "bytes": 1082411,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 27.483964940998703,
       "bytes": 11288612,
@@ -562,6 +556,12 @@
       "bytes": 6897324,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0080412499955855,
+      "bytes": 7466769,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -333,3 +333,4 @@
 {"step": 348300, "loss": 0.6221141424030066, "lr": 0.0003, "grad_norm": 0.17287579548104026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50982075928, "tokens_per_second": 149183.308537209, "data_wait_s": 1.2797994713764638e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348400, "loss": 0.6143290922045708, "lr": 0.0003, "grad_norm": 0.15746351881410195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50996773018, "tokens_per_second": 149744.95028023515, "data_wait_s": 9.178002073895186e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 348500, "loss": 0.6066722739487886, "lr": 0.0003, "grad_norm": 0.13950159604908102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51011587370, "tokens_per_second": 150502.17881202782, "data_wait_s": 9.357994713354856e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 348600, "loss": 0.5889758747071028, "lr": 0.0003, "grad_norm": 0.14416036690160747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51026133629, "tokens_per_second": 147350.11049295284, "data_wait_s": 1.1524003639351577e-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:adbfd95f8911e66a4a6053f63024c8ff6e8f563e7e0acd8f1a65fec1fe3b123f
-size 714393
+oid sha256:e7774e2138dfdc8f47ab0373cb0a84924782fc9142483e851b9946de8212a368
+size 716511