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": 74974.25439064938,
-      "required_pages_per_second_upper_estimate": 0.0064985783500709875
+      "required_token_positions_per_second": 75651.08409907826,
+      "required_pages_per_second_upper_estimate": 0.006557244233788389
     },
     {
       "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": 9371.781798831173,
-      "required_pages_per_second_upper_estimate": 0.0037385438801783837
+      "required_token_positions_per_second": 9456.385512384782,
+      "required_pages_per_second_upper_estimate": 0.0037722935664531603
     },
     {
       "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": 18743.563597662345,
-      "required_pages_per_second_upper_estimate": 0.00433599356286651
+      "required_token_positions_per_second": 18912.771024769565,
+      "required_pages_per_second_upper_estimate": 0.004375136723178758
     },
     {
       "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": 28115.345396493518,
-      "required_pages_per_second_upper_estimate": 0.0017120770248293192
+      "required_token_positions_per_second": 28369.156537154347,
+      "required_pages_per_second_upper_estimate": 0.0017275327916514236
     },
     {
       "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": 28115.345396493518,
-      "required_pages_per_second_upper_estimate": 0.002164397116879359
+      "required_token_positions_per_second": 28369.156537154347,
+      "required_pages_per_second_upper_estimate": 0.0021839362010816345
     },
     {
       "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": 28115.345396493518,
-      "required_pages_per_second_upper_estimate": 0.3490984938164262
+      "required_token_positions_per_second": 28369.156537154347,
+      "required_pages_per_second_upper_estimate": 0.3522499787321895
     }
   ],
   "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": 0.10281747285165466,
-    "producer_p95_seconds": 0.2194177219971607,
-    "gpu_update_seconds": 0.9833775687297157,
-    "producer_batches_per_second": 9.725973341542861,
-    "gpu_batches_per_second": 1.0169034069912297,
-    "producer_headroom": 9.564304018136447,
+    "producer_batch_seconds": 0.3420003724373828,
+    "producer_p95_seconds": 0.26711333699859097,
+    "gpu_update_seconds": 0.97457955663187,
+    "producer_batches_per_second": 2.9239734239853528,
+    "gpu_batches_per_second": 1.026083497437585,
+    "producer_headroom": 2.849644723151016,
     "sustainable": true,
-    "buffer_seconds": 60.969409261242376,
+    "buffer_seconds": 60.42393251117594,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 567882.0919186954,
-    "upload_bytes_per_second": 5924103.6878956985,
+    "generation_bytes_per_second": 537316.9442717854,
+    "upload_bytes_per_second": 5876340.701007189,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.927138653615216,
+    "estimated_full_shards_per_hour": 115.29570814242527,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 6189230,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9038558969987207,
+      "bytes": 2334921,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -116,3 +116,4 @@
 {"step": 96600, "loss": 0.6495152413845062, "lr": 0.0003, "grad_norm": 0.14796350184121645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14135810183, "tokens_per_second": 148593.40359386528, "data_wait_s": 1.0016003216151148e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 96700, "loss": 0.6217202711850405, "lr": 0.0003, "grad_norm": 0.1321169649126378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14150494538, "tokens_per_second": 149845.24128580684, "data_wait_s": 7.6029973570257425e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 96800, "loss": 0.6322421133518219, "lr": 0.0003, "grad_norm": 0.13472791764495293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14165285197, "tokens_per_second": 150639.13281617544, "data_wait_s": 1.1216001439606771e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 96900, "loss": 0.6075041946023703, "lr": 0.0003, "grad_norm": 0.13910251532313908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14179974930, "tokens_per_second": 149388.64069912166, "data_wait_s": 1.5313999028876424e-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 sha256:82c59aa015c321dbb974bf07f3ba081c19628cd1d43f979a0feb5888631b71ba
-size 251655
+oid sha256:f07f8f0aa3e22877ce3b684d057f477f3e1e98837a5d935599c9c45790413319
+size 253773