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": 75698.87529755497,
-      "required_pages_per_second_upper_estimate": 0.005237023838674783
+      "required_token_positions_per_second": 75805.700790147,
+      "required_pages_per_second_upper_estimate": 0.005244414274121593
     },
     {
       "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": 9462.359412194372,
-      "required_pages_per_second_upper_estimate": 0.003961692289351806
+      "required_token_positions_per_second": 9475.712598768376,
+      "required_pages_per_second_upper_estimate": 0.003967282989724097
     },
     {
       "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": 18924.718824388743,
-      "required_pages_per_second_upper_estimate": 0.003941519852723207
+      "required_token_positions_per_second": 18951.42519753675,
+      "required_pages_per_second_upper_estimate": 0.003947082085955516
     },
     {
       "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": 28387.078236583115,
-      "required_pages_per_second_upper_estimate": 0.0017017619957244217
+      "required_token_positions_per_second": 28427.137796305127,
+      "required_pages_per_second_upper_estimate": 0.0017041635051623504
     },
     {
       "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": 28387.078236583115,
-      "required_pages_per_second_upper_estimate": 0.002206970471332951
+      "required_token_positions_per_second": 28427.137796305127,
+      "required_pages_per_second_upper_estimate": 0.002210084925897956
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28387.078236583115,
-      "required_pages_per_second_upper_estimate": 0.3641189599489888
+      "required_token_positions_per_second": 28427.137796305127,
+      "required_pages_per_second_upper_estimate": 0.3646328009684987
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.43281881867193306,
-    "producer_p95_seconds": 0.25115406599798007,
-    "gpu_update_seconds": 0.9739642723909977,
-    "producer_batches_per_second": 2.3104355838048196,
-    "gpu_batches_per_second": 1.0267317070523407,
-    "producer_headroom": 2.250281712286731,
+    "producer_batch_seconds": 0.0674719651249518,
+    "producer_p95_seconds": 0.21915300999899046,
+    "gpu_update_seconds": 0.9725917606658804,
+    "producer_batches_per_second": 14.820970430431263,
+    "gpu_batches_per_second": 1.0281806205260824,
+    "producer_headroom": 14.414753725710094,
     "sustainable": true,
-    "buffer_seconds": 60.38578488824186,
+    "buffer_seconds": 60.30068916128458,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350431.9243617246,
-    "upload_bytes_per_second": 3825939.046495481,
+    "generation_bytes_per_second": 359854.78960337705,
+    "upload_bytes_per_second": 3899584.6157445074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.19453333033374,
+    "estimated_full_shards_per_hour": 77.216460858116,
     "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": 1.423224260999632,
-      "bytes": 2660337,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.6043403389994637,
       "bytes": 6329322,
@@ -562,6 +556,12 @@
       "bytes": 7013425,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.719156404997193,
+      "bytes": 11941861,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -165,3 +165,4 @@
 {"step": 816500, "loss": 0.5110582709312439, "lr": 3e-05, "grad_norm": 0.2823801772473911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119512226158, "tokens_per_second": 149414.69378761944, "data_wait_s": 1.541700112284161e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 816600, "loss": 0.5242553353309631, "lr": 3e-05, "grad_norm": 0.281920959251156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119526832504, "tokens_per_second": 151121.00753663926, "data_wait_s": 1.315700137638487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 816700, "loss": 0.5188537836074829, "lr": 3e-05, "grad_norm": 0.2697688917024653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119541401067, "tokens_per_second": 150342.95871078558, "data_wait_s": 6.4069972722791135e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 816800, "loss": 0.5265446305274963, "lr": 3e-05, "grad_norm": 0.27888402913266097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119556276816, "tokens_per_second": 153480.51494788963, "data_wait_s": 9.121002221945673e-06, "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 sha256:582f79723c3294def6a7f46fc6eeca4ee93b4a96de3455a8248c8c9928a46475
-size 356220
+oid sha256:0f9ad5bc3ec3fbe5d9b638e169e5c5ef3629053aae16beca8c6739b8eb08f195
+size 358338