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": 75637.69536132636,
-      "required_pages_per_second_upper_estimate": 0.006556083731933501
+      "required_token_positions_per_second": 75649.94788981916,
+      "required_pages_per_second_upper_estimate": 0.006557145750049506
     },
     {
       "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": 9454.711920165795,
-      "required_pages_per_second_upper_estimate": 0.0037716259454945726
+      "required_token_positions_per_second": 9456.243486227395,
+      "required_pages_per_second_upper_estimate": 0.003772236910095498
     },
     {
       "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": 18909.42384033159,
-      "required_pages_per_second_upper_estimate": 0.004374362410967467
+      "required_token_positions_per_second": 18912.48697245479,
+      "required_pages_per_second_upper_estimate": 0.004375071012674885
     },
     {
       "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": 28364.135760497382,
-      "required_pages_per_second_upper_estimate": 0.0017272270526949936
+      "required_token_positions_per_second": 28368.730458682185,
+      "required_pages_per_second_upper_estimate": 0.0017275068457079299
     },
     {
       "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": 28364.135760497382,
-      "required_pages_per_second_upper_estimate": 0.002183549687796182
+      "required_token_positions_per_second": 28368.730458682185,
+      "required_pages_per_second_upper_estimate": 0.002183903400381387
     },
     {
       "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": 28364.135760497382,
-      "required_pages_per_second_upper_estimate": 0.35218763748956855
+      "required_token_positions_per_second": 28368.730458682185,
+      "required_pages_per_second_upper_estimate": 0.35224468826355815
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1643137062970368,
-    "producer_p95_seconds": 0.41241809699567966,
-    "gpu_update_seconds": 0.9747520683674508,
-    "producer_batches_per_second": 6.085919565299428,
-    "gpu_batches_per_second": 1.0259019010596564,
-    "producer_headroom": 5.932262684193554,
+    "producer_batch_seconds": 0.10017717783620128,
+    "producer_p95_seconds": 0.39833074700436555,
+    "gpu_update_seconds": 0.9745941941345632,
+    "producer_batches_per_second": 9.982313552844243,
+    "gpu_batches_per_second": 1.0260680866132157,
+    "producer_headroom": 9.728704832632763,
     "sustainable": true,
-    "buffer_seconds": 60.43462823878195,
+    "buffer_seconds": 60.424840036342914,
     "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": 254569.13765190306,
-    "upload_bytes_per_second": 5973007.604793909,
+    "generation_bytes_per_second": 273845.7098428964,
+    "upload_bytes_per_second": 5900451.891907174,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.62461087386912,
+    "estimated_full_shards_per_hour": 58.760914530421914,
     "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": 0.8082868439996673,
-      "bytes": 1195987,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.157742789000622,
       "bytes": 11040923,
@@ -562,6 +556,12 @@
       "bytes": 7457751,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.6118608709948603,
+      "bytes": 885264,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -371,3 +371,4 @@
 {"step": 152100, "loss": 0.5953622683882713, "lr": 0.0003, "grad_norm": 0.13832643536771577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22259785078, "tokens_per_second": 149859.25446274687, "data_wait_s": 1.4823999663349241e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152200, "loss": 0.567689910531044, "lr": 0.0003, "grad_norm": 0.13095856199582015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22274231802, "tokens_per_second": 146985.4576452147, "data_wait_s": 1.0063995432574302e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152300, "loss": 0.61142473295331, "lr": 0.0003, "grad_norm": 0.13135204256711755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22288766647, "tokens_per_second": 148272.28295995138, "data_wait_s": 1.275399699807167e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 152400, "loss": 0.5488371104001999, "lr": 0.0003, "grad_norm": 0.13606638197257243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22303225480, "tokens_per_second": 96123.30438182088, "data_wait_s": 1.4270997780840844e-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:d6641cf3d868395158b9ffea4fba9b82e5804677ec2bb2b24110a8e874589551
-size 795660
+oid sha256:42b6d9ea171cc044201e96be24fa95debe8fbd37987278181b6b66bab2e91ad9
+size 797778