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": 77201.67287745554,
-      "required_pages_per_second_upper_estimate": 0.0053409908622231005
+      "required_token_positions_per_second": 76738.49066201359,
+      "required_pages_per_second_upper_estimate": 0.005308946841828017
     },
     {
       "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": 9650.209109681942,
-      "required_pages_per_second_upper_estimate": 0.004040341034942097
+      "required_token_positions_per_second": 9592.311332751699,
+      "required_pages_per_second_upper_estimate": 0.004016100444784472
     },
     {
       "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": 19300.418219363884,
-      "required_pages_per_second_upper_estimate": 0.004019768128837209
+      "required_token_positions_per_second": 19184.622665503397,
+      "required_pages_per_second_upper_estimate": 0.00399565096870218
     },
     {
       "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": 28950.627329045827,
-      "required_pages_per_second_upper_estimate": 0.0017355459033279212
+      "required_token_positions_per_second": 28776.933998255092,
+      "required_pages_per_second_upper_estimate": 0.0017251332533613764
     },
     {
       "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": 28950.627329045827,
-      "required_pages_per_second_upper_estimate": 0.002250783934481436
+      "required_token_positions_per_second": 28776.933998255092,
+      "required_pages_per_second_upper_estimate": 0.002237280042008677
     },
     {
       "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": 28950.627329045827,
-      "required_pages_per_second_upper_estimate": 0.37134756261522844
+      "required_token_positions_per_second": 28776.933998255092,
+      "required_pages_per_second_upper_estimate": 0.3691196110652133
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4964218595780494,
-    "producer_p95_seconds": 0.34940056299819844,
-    "gpu_update_seconds": 0.9550052123485796,
-    "producer_batches_per_second": 2.014415724662052,
-    "gpu_batches_per_second": 1.04711470374153,
-    "producer_headroom": 1.9237775168892013,
+    "producer_batch_seconds": 0.09341666077352784,
+    "producer_p95_seconds": 0.3257479259991669,
+    "gpu_update_seconds": 0.9607694830059538,
+    "producer_batches_per_second": 10.704728596800553,
+    "gpu_batches_per_second": 1.0408323928767034,
+    "producer_headroom": 10.284776559667117,
     "sustainable": true,
-    "buffer_seconds": 60.16532837796051,
+    "buffer_seconds": 60.52847742937509,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292058.9534759953,
-    "upload_bytes_per_second": 4236218.212568838,
+    "generation_bytes_per_second": 292092.2769028455,
+    "upload_bytes_per_second": 4292457.8828252815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.66905263147253,
+    "estimated_full_shards_per_hour": 62.676203063144904,
     "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": 1.5667999549987144,
-      "bytes": 6923893,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.4826895270016394,
       "bytes": 6163509,
@@ -562,6 +556,12 @@
       "bytes": 2146631,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.2327626959959161,
+      "bytes": 11052547,
+      "pages": 48,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -408,3 +408,4 @@
 {"step": 840800, "loss": 0.5980193018913269, "lr": 3e-05, "grad_norm": 0.32692752093060673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123069241731, "tokens_per_second": 150399.62103053424, "data_wait_s": 1.2342003174126148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840900, "loss": 0.5097891688346863, "lr": 3e-05, "grad_norm": 0.28425541977736746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123083732760, "tokens_per_second": 133410.07703303223, "data_wait_s": 9.864997991826385e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 841000, "loss": 0.5003107190132141, "lr": 3e-05, "grad_norm": 0.3166021561959973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123098235734, "tokens_per_second": 150518.3236412449, "data_wait_s": 9.83799691312015e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 841100, "loss": 0.5524387359619141, "lr": 3e-05, "grad_norm": 0.30640263690148817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123112889526, "tokens_per_second": 151475.88977120374, "data_wait_s": 8.356000762432814e-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 sha256:84b36f598b4a1024cca1eac2a036e7a441b73e92f4ca8807c6e9bae01bf48bbb
-size 874809
+oid sha256:d4fa10aa7d43fb588303fd5d557225a4f98c3223c7632244c084594c6e0f64c5
+size 876927