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": 77029.1976194062,
-      "required_pages_per_second_upper_estimate": 0.005329058623673512
+      "required_token_positions_per_second": 77817.09611100072,
+      "required_pages_per_second_upper_estimate": 0.005383567269498395
     },
     {
       "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": 9628.649702425775,
-      "required_pages_per_second_upper_estimate": 0.004031314561335559
+      "required_token_positions_per_second": 9727.13701387509,
+      "required_pages_per_second_upper_estimate": 0.004072549141990455
     },
     {
       "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": 19257.29940485155,
-      "required_pages_per_second_upper_estimate": 0.004010787616893897
+      "required_token_positions_per_second": 19454.27402775018,
+      "required_pages_per_second_upper_estimate": 0.0040518122362735555
     },
     {
       "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": 28885.949107277323,
-      "required_pages_per_second_upper_estimate": 0.0017316685426908259
+      "required_token_positions_per_second": 29181.411041625266,
+      "required_pages_per_second_upper_estimate": 0.0017493810345107334
     },
     {
       "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": 28885.949107277323,
-      "required_pages_per_second_upper_estimate": 0.0022457554872283665
+      "required_token_positions_per_second": 29181.411041625266,
+      "required_pages_per_second_upper_estimate": 0.0022687263530242158
     },
     {
       "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": 28885.949107277323,
-      "required_pages_per_second_upper_estimate": 0.370517939832446
+      "required_token_positions_per_second": 29181.411041625266,
+      "required_pages_per_second_upper_estimate": 0.3743078082839531
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04516078764064346,
-    "producer_p95_seconds": 0.14055661700012934,
-    "gpu_update_seconds": 0.9571435543737961,
-    "producer_batches_per_second": 22.143103613632007,
-    "gpu_batches_per_second": 1.0447753583361301,
-    "producer_headroom": 21.194128897618988,
+    "producer_batch_seconds": 0.20642152302344385,
+    "producer_p95_seconds": 0.14950838300001124,
+    "gpu_update_seconds": 0.9474524710461066,
+    "producer_batches_per_second": 4.84445606908165,
+    "gpu_batches_per_second": 1.0554619155680436,
+    "producer_headroom": 4.589891873525717,
     "sustainable": true,
-    "buffer_seconds": 60.300043925549154,
+    "buffer_seconds": 60.636958146950825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 489330.47618760617,
-    "upload_bytes_per_second": 4189765.118786665,
+    "generation_bytes_per_second": 481130.0074377952,
+    "upload_bytes_per_second": 4351065.513279547,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.99892915936603,
+    "estimated_full_shards_per_hour": 103.239299462799,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 6239919,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9808759570000802,
+      "bytes": 5815079,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -20,3 +20,4 @@
 {"step": 887000, "loss": 0.5029613971710205, "lr": 3e-05, "grad_norm": 0.3043762816644715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129832258644, "tokens_per_second": 154174.97356981126, "data_wait_s": 1.4352999642142095e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887100, "loss": 0.5219621062278748, "lr": 3e-05, "grad_norm": 0.34227147279043973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129847050100, "tokens_per_second": 153810.14400547234, "data_wait_s": 1.4380000266100978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887200, "loss": 0.5042001605033875, "lr": 3e-05, "grad_norm": 0.3097639490348491, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129861565452, "tokens_per_second": 152243.22139336707, "data_wait_s": 1.0821000159921823e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 887300, "loss": 0.541535496711731, "lr": 3e-05, "grad_norm": 0.30549372524149293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129876072345, "tokens_per_second": 148892.38539015697, "data_wait_s": 1.0383000244473806e-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 sha256:15a5bb1675b1947f96a483985d3bfbba8dd8be65f7a3105e1dbc1e521d4fae02
-size 46761
+oid sha256:6de8cfde14704332af27fa4b818568976d37ca64bcbfba6ac62153a7540e4b90
+size 48879