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": 77344.6543232231,
-      "required_pages_per_second_upper_estimate": 0.005350882650403963
+      "required_token_positions_per_second": 77289.7429755001,
+      "required_pages_per_second_upper_estimate": 0.005347083755956605
     },
     {
       "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": 9668.081790402888,
-      "required_pages_per_second_upper_estimate": 0.004047823953135943
+      "required_token_positions_per_second": 9661.217871937513,
+      "required_pages_per_second_upper_estimate": 0.004044950173809408
     },
     {
       "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": 19336.163580805776,
-      "required_pages_per_second_upper_estimate": 0.004027212944956982
+      "required_token_positions_per_second": 19322.435743875027,
+      "required_pages_per_second_upper_estimate": 0.004024353798551712
     },
     {
       "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": 29004.245371208664,
-      "required_pages_per_second_upper_estimate": 0.0017387602230855735
+      "required_token_positions_per_second": 28983.65361581254,
+      "required_pages_per_second_upper_estimate": 0.0017375257787913656
     },
     {
       "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": 29004.245371208664,
-      "required_pages_per_second_upper_estimate": 0.0022549525014187535
+      "required_token_positions_per_second": 28983.65361581254,
+      "required_pages_per_second_upper_estimate": 0.0022533515830102644
     },
     {
       "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": 29004.245371208664,
-      "required_pages_per_second_upper_estimate": 0.3720353172895251
+      "required_token_positions_per_second": 28983.65361581254,
+      "required_pages_per_second_upper_estimate": 0.3717711883610079
     }
   ],
   "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": 0.0751649298984205,
-    "producer_p95_seconds": 0.15723417199933465,
-    "gpu_update_seconds": 0.9532397635638901,
-    "producer_batches_per_second": 13.304076799531664,
-    "gpu_batches_per_second": 1.0490540137155913,
-    "producer_headroom": 12.681975022821398,
+    "producer_batch_seconds": 0.07206851687498528,
+    "producer_p95_seconds": 0.15982047600027727,
+    "gpu_update_seconds": 0.9539170032350976,
+    "producer_batches_per_second": 13.875684464752686,
+    "gpu_batches_per_second": 1.0483092308959976,
+    "producer_headroom": 13.236251342452682,
     "sustainable": true,
-    "buffer_seconds": 60.05410510452507,
+    "buffer_seconds": 60.09677120381115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369492.52396627737,
-    "upload_bytes_per_second": 5176279.881902212,
+    "generation_bytes_per_second": 369666.6439497232,
+    "upload_bytes_per_second": 5020190.477289464,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.28449429742089,
+    "estimated_full_shards_per_hour": 79.32185639256261,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 16176474,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.33469606700055,
+      "bytes": 2416191,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 888800, "loss": 0.4857497215270996, "lr": 3e-05, "grad_norm": 0.32665121437354316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130095710575, "tokens_per_second": 138544.80322139457, "data_wait_s": 9.969000529963523e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 888900, "loss": 0.5321134924888611, "lr": 3e-05, "grad_norm": 0.31782447520592444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130110243844, "tokens_per_second": 151452.7059806842, "data_wait_s": 1.4088000170886517e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889000, "loss": 0.4985983073711395, "lr": 3e-05, "grad_norm": 0.312134335490036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130124948652, "tokens_per_second": 153313.03883898287, "data_wait_s": 2.075199972750852e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 889100, "loss": 0.5271111130714417, "lr": 3e-05, "grad_norm": 0.3237032503284707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130139706737, "tokens_per_second": 152814.1421765432, "data_wait_s": 1.918299949466018e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79e1f23037a03efb4f14698b3bac6f5295158d7fff20772e680bb05ea633151a
-size 84885
+oid sha256:1d1b3891fa7435906e738448f3d9c71c2d5dce0ae04a1472ef571c44ed0dca4e
+size 87003