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": 75192.25529963699,
-      "required_pages_per_second_upper_estimate": 0.006517474116343781
+      "required_token_positions_per_second": 75733.47975957241,
+      "required_pages_per_second_upper_estimate": 0.00656438608081013
     },
     {
       "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": 9399.031912454624,
-      "required_pages_per_second_upper_estimate": 0.0037494143579282846
+      "required_token_positions_per_second": 9466.684969946551,
+      "required_pages_per_second_upper_estimate": 0.003776402174065163
     },
     {
       "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": 18798.063824909248,
-      "required_pages_per_second_upper_estimate": 0.004348601231268841
+      "required_token_positions_per_second": 18933.369939893102,
+      "required_pages_per_second_upper_estimate": 0.004379901919664064
     },
     {
       "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": 28197.09573736387,
-      "required_pages_per_second_upper_estimate": 0.0017170551916774331
+      "required_token_positions_per_second": 28400.05490983965,
+      "required_pages_per_second_upper_estimate": 0.0017294143404367258
     },
     {
       "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": 28197.09573736387,
-      "required_pages_per_second_upper_estimate": 0.0021706904844189857
+      "required_token_positions_per_second": 28400.05490983965,
+      "required_pages_per_second_upper_estimate": 0.0021863148433431193
     },
     {
       "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": 28197.09573736387,
-      "required_pages_per_second_upper_estimate": 0.3501135594492453
+      "required_token_positions_per_second": 28400.05490983965,
+      "required_pages_per_second_upper_estimate": 0.3526336331107398
     }
   ],
   "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": 0.0767356367891523,
-    "producer_p95_seconds": 0.1991816280024068,
-    "gpu_update_seconds": 0.9805265144155869,
-    "producer_batches_per_second": 13.031754760147695,
-    "gpu_batches_per_second": 1.0198602335562743,
-    "producer_headroom": 12.777981071686352,
+    "producer_batch_seconds": 0.20705526900795235,
+    "producer_p95_seconds": 0.26620429800095735,
+    "gpu_update_seconds": 0.973519244514591,
+    "producer_batches_per_second": 4.829628363437557,
+    "gpu_batches_per_second": 1.0272010601070476,
+    "producer_headroom": 4.7017361556599715,
     "sustainable": true,
-    "buffer_seconds": 60.79264389376639,
+    "buffer_seconds": 60.358193159904644,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481718.1072471379,
-    "upload_bytes_per_second": 5846447.101302481,
+    "generation_bytes_per_second": 492270.96945209627,
+    "upload_bytes_per_second": 5845659.336047131,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.36549199162104,
+    "estimated_full_shards_per_hour": 105.62989056274573,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 1118965,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.006351252002787,
+      "bytes": 5851026,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -92,3 +92,4 @@
 {"step": 94200, "loss": 0.6251903437077999, "lr": 0.0003, "grad_norm": 0.1315070496077938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13785653862, "tokens_per_second": 148819.15678524968, "data_wait_s": 6.8160006776452065e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94300, "loss": 0.6212357766926289, "lr": 0.0003, "grad_norm": 0.13607407178389822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13800147761, "tokens_per_second": 147773.20663485298, "data_wait_s": 9.108000085689127e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94400, "loss": 0.6433022990822792, "lr": 0.0003, "grad_norm": 0.13999482436726243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13814687815, "tokens_per_second": 148644.52013275062, "data_wait_s": 9.969000529963523e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 94500, "loss": 0.6318197883665562, "lr": 0.0003, "grad_norm": 0.15126675226290945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13829351755, "tokens_per_second": 149767.10950509724, "data_wait_s": 1.0874999134102836e-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 sha256:18796db0ef743441c1bede48c9f138d92fdf240f86836f62bb03e03b90925ff4
-size 200040
+oid sha256:70cda0b97281d0004bf27bc0f04726bd2c51f7889f6549970c3d6621551d78a2
+size 202158