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": 76854.15556388156,
-      "required_pages_per_second_upper_estimate": 0.005316948808118792
+      "required_token_positions_per_second": 76711.87345545282,
+      "required_pages_per_second_upper_estimate": 0.005307105401717715
     },
     {
       "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": 9606.769445485195,
-      "required_pages_per_second_upper_estimate": 0.004022153754666261
+      "required_token_positions_per_second": 9588.984181931603,
+      "required_pages_per_second_upper_estimate": 0.004014707436214907
     },
     {
       "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": 19213.53889097039,
-      "required_pages_per_second_upper_estimate": 0.004001673455894808
+      "required_token_positions_per_second": 19177.968363863205,
+      "required_pages_per_second_upper_estimate": 0.003994265053156263
     },
     {
       "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": 28820.308336455586,
-      "required_pages_per_second_upper_estimate": 0.0017277334787077162
+      "required_token_positions_per_second": 28766.952545794808,
+      "required_pages_per_second_upper_estimate": 0.0017245348805271777
     },
     {
       "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": 28820.308336455586,
-      "required_pages_per_second_upper_estimate": 0.002240652206020214
+      "required_token_positions_per_second": 28766.952545794808,
+      "required_pages_per_second_upper_estimate": 0.002236504027983659
     },
     {
       "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": 28820.308336455586,
-      "required_pages_per_second_upper_estimate": 0.3696759705039133
+      "required_token_positions_per_second": 28766.952545794808,
+      "required_pages_per_second_upper_estimate": 0.3689915797103014
     }
   ],
   "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": 0.1298943250938862,
-    "producer_p95_seconds": 0.3049356090014044,
-    "gpu_update_seconds": 0.9593235324629508,
-    "producer_batches_per_second": 7.698565732392164,
-    "gpu_batches_per_second": 1.0424011985118484,
-    "producer_headroom": 7.385415273296675,
+    "producer_batch_seconds": 0.08327035632015622,
+    "producer_p95_seconds": 0.28712650899979053,
+    "gpu_update_seconds": 0.9611028473032199,
+    "producer_batches_per_second": 12.00907554850876,
+    "gpu_batches_per_second": 1.0404713739075089,
+    "producer_headroom": 11.541956703151246,
     "sustainable": true,
-    "buffer_seconds": 60.437382545165896,
+    "buffer_seconds": 60.54947938010285,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400355.97463445854,
-    "upload_bytes_per_second": 5031071.611693859,
+    "generation_bytes_per_second": 400583.3318873306,
+    "upload_bytes_per_second": 4968058.456927287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.90707234645193,
+    "estimated_full_shards_per_hour": 85.9558579203123,
     "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": 1.0737022979992616,
-      "bytes": 6591396,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.1665754960013146,
       "bytes": 11803743,
@@ -562,6 +556,12 @@
       "bytes": 11699991,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0364311270022881,
+      "bytes": 982135,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 590000, "loss": 0.5705015063285828, "lr": 0.0003, "grad_norm": 0.1529459236000265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86358244059, "tokens_per_second": 152205.71142108247, "data_wait_s": 1.3642998965224251e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590100, "loss": 0.6136918067932129, "lr": 0.0003, "grad_norm": 0.14434504991304872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86372899538, "tokens_per_second": 138442.23028971525, "data_wait_s": 8.31099896458909e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 590200, "loss": 0.5751941204071045, "lr": 0.0003, "grad_norm": 0.16299431189703864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86387570881, "tokens_per_second": 152453.57019604737, "data_wait_s": 1.2955999409314245e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 590300, "loss": 0.5869466066360474, "lr": 0.0003, "grad_norm": 0.1953576552185063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86402190538, "tokens_per_second": 152258.43085131663, "data_wait_s": 1.1400999937904999e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb8777e49e5cce91e437f7674e019b2d18961522edb1f82af7f56ed6324fb864
-size 706704
+oid sha256:8086e0c3b6035c4071268b8dc64f68c9a28de6005898cd58faf1177b2bd80c9f
+size 708822