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": 76004.11313910095,
-      "required_pages_per_second_upper_estimate": 0.0065878438962336345
+      "required_token_positions_per_second": 76066.70311640597,
+      "required_pages_per_second_upper_estimate": 0.006593269036833587
     },
     {
       "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": 9500.51414238762,
-      "required_pages_per_second_upper_estimate": 0.0037898971367431064
+      "required_token_positions_per_second": 9508.337889550747,
+      "required_pages_per_second_upper_estimate": 0.0037930181464619224
     },
     {
       "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": 19001.02828477524,
-      "required_pages_per_second_upper_estimate": 0.004395553487006346
+      "required_token_positions_per_second": 19016.675779101493,
+      "required_pages_per_second_upper_estimate": 0.00439917325943223
     },
     {
       "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": 28501.542427162858,
-      "required_pages_per_second_upper_estimate": 0.0017355943977778565
+      "required_token_positions_per_second": 28525.01366865224,
+      "required_pages_per_second_upper_estimate": 0.0017370236732404728
     },
     {
       "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": 28501.542427162858,
-      "required_pages_per_second_upper_estimate": 0.002194127633362088
+      "required_token_positions_per_second": 28525.01366865224,
+      "required_pages_per_second_upper_estimate": 0.0021959345145044955
     },
     {
       "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": 28501.542427162858,
-      "required_pages_per_second_upper_estimate": 0.35389376841902304
+      "required_token_positions_per_second": 28525.01366865224,
+      "required_pages_per_second_upper_estimate": 0.35418520268512904
     }
   ],
   "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": 0.08777022086701436,
-    "producer_p95_seconds": 0.344478678998712,
-    "gpu_update_seconds": 0.9700527636585238,
-    "producer_batches_per_second": 11.393385935705423,
-    "gpu_batches_per_second": 1.0308717602417121,
-    "producer_headroom": 11.0521855143592,
+    "producer_batch_seconds": 0.10863376663260738,
+    "producer_p95_seconds": 0.3543557859993598,
+    "gpu_update_seconds": 0.969254574990229,
+    "producer_batches_per_second": 9.205240976150055,
+    "gpu_batches_per_second": 1.031720691140489,
+    "producer_headroom": 8.922221930020962,
     "sustainable": true,
-    "buffer_seconds": 60.14327134682848,
+    "buffer_seconds": 60.0937836493942,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432198.7760998825,
-    "upload_bytes_per_second": 3939755.313660088,
+    "generation_bytes_per_second": 431870.0562857801,
+    "upload_bytes_per_second": 4019365.477789995,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.73979627845151,
+    "estimated_full_shards_per_hour": 92.66926065855075,
     "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": 1.6848288789988146,
-      "bytes": 823777,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.0912059820002469,
       "bytes": 11909059,
@@ -562,6 +556,12 @@
       "bytes": 5850558,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8959461580016068,
+      "bytes": 6085583,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -324,3 +324,4 @@
 {"step": 72400, "loss": 0.6408884692937136, "lr": 0.0003, "grad_norm": 0.15654120584981696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10595843036, "tokens_per_second": 147119.27267972243, "data_wait_s": 1.3313001545611769e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 72500, "loss": 0.6018617432564497, "lr": 0.0003, "grad_norm": 0.13016087217758465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10610476137, "tokens_per_second": 149884.14301566096, "data_wait_s": 1.312200038228184e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 72600, "loss": 0.6149622984230518, "lr": 0.0003, "grad_norm": 0.13127798112473363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10625047722, "tokens_per_second": 149422.48085558575, "data_wait_s": 1.017199974739924e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 72700, "loss": 0.6209270842373371, "lr": 0.0003, "grad_norm": 0.13544607122665395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10639721542, "tokens_per_second": 150185.50780856414, "data_wait_s": 1.168000017059967e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c4ced6121a8aa166bb032591106507edd31d889211853de1bdf6e6831cf0e91
-size 695331
+oid sha256:51e36355232d72d673586834f65d5b601f8c64cae40ed0e298f1a52035d943ad
+size 697449