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": 75559.00860847537,
-      "required_pages_per_second_upper_estimate": 0.006549263363626129
+      "required_token_positions_per_second": 75169.66499260702,
+      "required_pages_per_second_upper_estimate": 0.006515516045785033
     },
     {
       "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": 9444.876076059421,
-      "required_pages_per_second_upper_estimate": 0.003767702280221566
+      "required_token_positions_per_second": 9396.208124075878,
+      "required_pages_per_second_upper_estimate": 0.0037482879065246044
     },
     {
       "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": 18889.752152118843,
-      "required_pages_per_second_upper_estimate": 0.004369811712109335
+      "required_token_positions_per_second": 18792.416248151756,
+      "required_pages_per_second_upper_estimate": 0.004347294763779952
     },
     {
       "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": 28334.628228178262,
-      "required_pages_per_second_upper_estimate": 0.0017254301987908167
+      "required_token_positions_per_second": 28188.624372227638,
+      "required_pages_per_second_upper_estimate": 0.0017165393299864544
     },
     {
       "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": 28334.628228178262,
-      "required_pages_per_second_upper_estimate": 0.002181278116276179
+      "required_token_positions_per_second": 28188.624372227638,
+      "required_pages_per_second_upper_estimate": 0.002170038335280559
     },
     {
       "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": 28334.628228178262,
-      "required_pages_per_second_upper_estimate": 0.351821252693523
+      "required_token_positions_per_second": 28188.624372227638,
+      "required_pages_per_second_upper_estimate": 0.3500083734460886
     }
   ],
   "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": 0.19624237472669392,
-    "producer_p95_seconds": 0.3505261579994112,
-    "gpu_update_seconds": 0.975767170027824,
-    "producer_batches_per_second": 5.095739395697267,
-    "gpu_batches_per_second": 1.024834643669642,
-    "producer_headroom": 4.972255209338816,
+    "producer_batch_seconds": 0.15410881240626395,
+    "producer_p95_seconds": 0.3714157790018362,
+    "gpu_update_seconds": 0.9808211864087885,
+    "producer_batches_per_second": 6.488921589790628,
+    "gpu_batches_per_second": 1.0195538329075389,
+    "producer_headroom": 6.364471772212046,
     "sustainable": true,
-    "buffer_seconds": 60.49756454172509,
+    "buffer_seconds": 60.81091355734488,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491554.28808109014,
+    "generation_bytes_per_second": 486230.2694043598,
     "upload_bytes_per_second": 6002354.254787021,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.47610742401626,
+    "estimated_full_shards_per_hour": 104.33369695280166,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 114300, "loss": 0.6063061691820621, "lr": 0.0003, "grad_norm": 0.1685130957539429, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16725834847, "tokens_per_second": 148726.3324610475, "data_wait_s": 1.841800258262083e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114400, "loss": 0.6041073054075241, "lr": 0.0003, "grad_norm": 0.15474263024281182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16740523488, "tokens_per_second": 150310.6691414158, "data_wait_s": 1.647500175749883e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114500, "loss": 0.6019764691591263, "lr": 0.0003, "grad_norm": 0.13809368823880785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16755176941, "tokens_per_second": 149165.51744536447, "data_wait_s": 2.0373001461848617e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 114600, "loss": 0.5923062432557344, "lr": 0.0003, "grad_norm": 0.13503638414743294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16769917267, "tokens_per_second": 150322.0356558141, "data_wait_s": 1.4290999388322234e-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 sha256:97ea84b6bdaa23ef1a08e8283a88fce02e93614dcb27f680d960bfb5391cbf44
-size 628890
+oid sha256:067e7d2254ff7a8d42f5c1f2483b067942813fee8b880ffd6f69354add098070
+size 631008