Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76593.56909363675,
-      "required_pages_per_second_upper_estimate": 0.00529892083146334
+      "required_token_positions_per_second": 76586.06278082573,
+      "required_pages_per_second_upper_estimate": 0.005298401527326029
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9574.196136704593,
-      "required_pages_per_second_upper_estimate": 0.004008515990487859
+      "required_token_positions_per_second": 9573.257847603216,
+      "required_pages_per_second_upper_estimate": 0.004008123148434816
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19148.392273409187,
-      "required_pages_per_second_upper_estimate": 0.003988105133488647
+      "required_token_positions_per_second": 19146.51569520643,
+      "required_pages_per_second_upper_estimate": 0.0039877142917377085
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28722.58841011378,
-      "required_pages_per_second_upper_estimate": 0.0017218753183332132
+      "required_token_positions_per_second": 28719.773542809646,
+      "required_pages_per_second_upper_estimate": 0.0017217065713363837
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28722.58841011378,
-      "required_pages_per_second_upper_estimate": 0.0022330549115716697
+      "required_token_positions_per_second": 28719.773542809646,
+      "required_pages_per_second_upper_estimate": 0.0022328360680200655
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28722.58841011378,
-      "required_pages_per_second_upper_estimate": 0.36842252421228283
+      "required_token_positions_per_second": 28719.773542809646,
+      "required_pages_per_second_upper_estimate": 0.3683864181168744
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08864033103906621,
-    "producer_p95_seconds": 0.1794391600005838,
-    "gpu_update_seconds": 0.9625873408492879,
-    "producer_batches_per_second": 11.281546315065912,
-    "gpu_batches_per_second": 1.03886676830562,
-    "producer_headroom": 10.859473668087379,
+    "producer_batch_seconds": 0.06701315362498406,
+    "producer_p95_seconds": 0.1680038690001311,
+    "gpu_update_seconds": 0.9626816854522874,
+    "producer_batches_per_second": 14.922443519016493,
+    "gpu_batches_per_second": 1.0387649574222237,
+    "producer_headroom": 14.365563077953361,
     "sustainable": true,
-    "buffer_seconds": 60.643002473505135,
+    "buffer_seconds": 60.648946183494104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396191.56637706544,
-    "upload_bytes_per_second": 3799671.5982692074,
+    "generation_bytes_per_second": 416985.3448216347,
+    "upload_bytes_per_second": 3809992.7171074897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.0134872768781,
+    "estimated_full_shards_per_hour": 89.4753480766943,
     "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": [
@@ -274,6 +274,12 @@
       "bytes": 7689453,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.5363667999999961,
+      "bytes": 6120972,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -58,3 +58,4 @@
 {"step": 805800, "loss": 0.5090294480323792, "lr": 3e-05, "grad_norm": 0.26825289216626286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117945672353, "tokens_per_second": 149602.61917522838, "data_wait_s": 1.5653999980713706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805900, "loss": 0.5663489699363708, "lr": 3e-05, "grad_norm": 0.2863788143839216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117960273810, "tokens_per_second": 151720.84326337787, "data_wait_s": 1.4418999853660353e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806000, "loss": 0.5180491209030151, "lr": 3e-05, "grad_norm": 0.25843856561638007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117974916154, "tokens_per_second": 151817.38622671983, "data_wait_s": 1.479599995946046e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 806100, "loss": 0.5533642172813416, "lr": 3e-05, "grad_norm": 0.2787937548759742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117989609136, "tokens_per_second": 151631.99965050165, "data_wait_s": 1.6586999663559254e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f7744282fb56e86e5c08f4407932ff658d039bf2299bcb87118c6fbd202d92d
-size 128028
+oid sha256:a095528454af7343c0a470291a89d9ad2afe8874004e5a58f1c7c368ae23a075
+size 130146