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": 76845.6332959068,
-      "required_pages_per_second_upper_estimate": 0.005316359217845911
+      "required_token_positions_per_second": 76862.4212905593,
+      "required_pages_per_second_upper_estimate": 0.005317520650269486
     },
     {
       "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": 9605.70416198835,
-      "required_pages_per_second_upper_estimate": 0.00402170774271178
+      "required_token_positions_per_second": 9607.802661319913,
+      "required_pages_per_second_upper_estimate": 0.004022586340560257
     },
     {
       "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": 19211.4083239767,
-      "required_pages_per_second_upper_estimate": 0.004001229714976852
+      "required_token_positions_per_second": 19215.605322639825,
+      "required_pages_per_second_upper_estimate": 0.004002103839116084
     },
     {
       "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": 28817.11248596505,
-      "required_pages_per_second_upper_estimate": 0.0017275418923505899
+      "required_token_positions_per_second": 28823.40798395974,
+      "required_pages_per_second_upper_estimate": 0.0017279192978427026
     },
     {
       "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": 28817.11248596505,
-      "required_pages_per_second_upper_estimate": 0.0022404037427016356
+      "required_token_positions_per_second": 28823.40798395974,
+      "required_pages_per_second_upper_estimate": 0.0022408931899797536
     },
     {
       "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": 28817.11248596505,
-      "required_pages_per_second_upper_estimate": 0.36963497756525765
+      "required_token_positions_per_second": 28823.40798395974,
+      "required_pages_per_second_upper_estimate": 0.3697157294539544
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07758627153910425,
-    "producer_p95_seconds": 0.23995092199766077,
-    "gpu_update_seconds": 0.9594299225318134,
-    "producer_batches_per_second": 12.888878150253555,
-    "gpu_batches_per_second": 1.0422856078546388,
-    "producer_headroom": 12.36597536521975,
+    "producer_batch_seconds": 0.06686083496080641,
+    "producer_p95_seconds": 0.23434182499840972,
+    "gpu_update_seconds": 0.9592203675355164,
+    "producer_batches_per_second": 14.956439006276192,
+    "gpu_batches_per_second": 1.0425133096050252,
+    "producer_headroom": 14.346520920622783,
     "sustainable": true,
-    "buffer_seconds": 60.444085119504244,
+    "buffer_seconds": 60.430883154737536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435977.4238918228,
-    "upload_bytes_per_second": 5806384.255829415,
+    "generation_bytes_per_second": 435230.64047260483,
+    "upload_bytes_per_second": 5708077.823417961,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.5506061321832,
+    "estimated_full_shards_per_hour": 93.3903637946473,
     "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": 0.963736620000418,
-      "bytes": 7684006,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1074004389993206,
       "bytes": 6213956,
@@ -562,6 +556,12 @@
       "bytes": 11260844,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.0456256510005915,
+      "bytes": 862691,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -170,3 +170,4 @@
 {"step": 414100, "loss": 0.544400691986084, "lr": 0.0003, "grad_norm": 0.16984980498497781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60613406530, "tokens_per_second": 150581.64305898597, "data_wait_s": 1.7678998119663447e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 414200, "loss": 0.5877511501312256, "lr": 0.0003, "grad_norm": 0.17852342157700163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60627932470, "tokens_per_second": 150856.1027046889, "data_wait_s": 1.2816999515052885e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 414300, "loss": 0.5566993951797485, "lr": 0.0003, "grad_norm": 0.14471358873226575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60642563852, "tokens_per_second": 151106.99731864664, "data_wait_s": 1.5335001080529764e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 414400, "loss": 0.5966958999633789, "lr": 0.0003, "grad_norm": 0.15678041712656168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60657135585, "tokens_per_second": 138509.9481808908, "data_wait_s": 1.4477998774964362e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d489cceb8e5fd8fdae1c65825ea9cab131e6d968d3950b993bf114e53c7fe90a
-size 366810
+oid sha256:6b9619e8d786e3379c904b84305815241fbd2a1f638dbadc065cd65f51a7d742
+size 368928