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": 75592.8191743543,
-      "required_pages_per_second_upper_estimate": 0.006552193977784418
+      "required_token_positions_per_second": 75089.82802847157,
+      "required_pages_per_second_upper_estimate": 0.006508595980078708
     },
     {
       "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": 9449.102396794287,
-      "required_pages_per_second_upper_estimate": 0.00376938822275183
+      "required_token_positions_per_second": 9386.228503558947,
+      "required_pages_per_second_upper_estimate": 0.003744306886691777
     },
     {
       "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": 18898.204793588575,
-      "required_pages_per_second_upper_estimate": 0.004371767081951949
+      "required_token_positions_per_second": 18772.457007117893,
+      "required_pages_per_second_upper_estimate": 0.004342677544637413
     },
     {
       "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": 28347.30719038286,
-      "required_pages_per_second_upper_estimate": 0.001726202281067703
+      "required_token_positions_per_second": 28158.685510676838,
+      "required_pages_per_second_upper_estimate": 0.0017147162103950777
     },
     {
       "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": 28347.30719038286,
-      "required_pages_per_second_upper_estimate": 0.0021822541778843026
+      "required_token_positions_per_second": 28158.685510676838,
+      "required_pages_per_second_upper_estimate": 0.002167733558842305
     },
     {
       "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": 28347.30719038286,
-      "required_pages_per_second_upper_estimate": 0.3519786829703473
+      "required_token_positions_per_second": 28158.685510676838,
+      "required_pages_per_second_upper_estimate": 0.34963663298455167
     }
   ],
   "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": 0.0684410804062523,
-    "producer_p95_seconds": 0.18613033600013296,
-    "gpu_update_seconds": 0.9753307365074836,
-    "producer_batches_per_second": 14.611107745000574,
-    "gpu_batches_per_second": 1.0252932288188246,
-    "producer_headroom": 14.250662478121608,
+    "producer_batch_seconds": 0.08059340783588453,
+    "producer_p95_seconds": 0.16903670000010607,
+    "gpu_update_seconds": 0.9818640145512757,
+    "producer_batches_per_second": 12.407962721173655,
+    "gpu_batches_per_second": 1.0184709747785314,
+    "producer_headroom": 12.182932089814136,
     "sustainable": true,
-    "buffer_seconds": 60.47050566346398,
+    "buffer_seconds": 60.87556890217909,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 534753.2273098361,
-    "upload_bytes_per_second": 5723910.322731022,
+    "generation_bytes_per_second": 551638.3017967462,
+    "upload_bytes_per_second": 5754555.121828018,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.74559416266739,
+    "estimated_full_shards_per_hour": 118.36873808314122,
     "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": [
@@ -382,6 +382,12 @@
       "bytes": 967616,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 0.879939514000398,
+      "bytes": 6151463,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 85700, "loss": 0.6051355823874474, "lr": 0.0003, "grad_norm": 0.1315521462717127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12541770395, "tokens_per_second": 148511.03501026588, "data_wait_s": 1.2501999663072638e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 85800, "loss": 0.5889550969004631, "lr": 0.0003, "grad_norm": 0.16294887114772708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12556535738, "tokens_per_second": 149503.7780028432, "data_wait_s": 9.805999980017077e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 85900, "loss": 0.6246379911899567, "lr": 0.0003, "grad_norm": 0.13538628748801487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12571121491, "tokens_per_second": 147658.76700833597, "data_wait_s": 8.027000149013475e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 86000, "loss": 0.5379970334470272, "lr": 0.0003, "grad_norm": 0.13144490846414586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12585673405, "tokens_per_second": 148034.74707902598, "data_wait_s": 1.1390000508981757e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f3713fbf8228dcbea3a0c19ef96620742df93cb37d2ac8bfc74d0db95f6db5d
-size 85668
+oid sha256:098341505e239096220ec547a31af4daa8c028bfccc90985942323827aa6128f
+size 87786