Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75371.6302428426,
-      "required_pages_per_second_upper_estimate": 0.006533021881799206
+      "required_token_positions_per_second": 75559.51652648897,
+      "required_pages_per_second_upper_estimate": 0.00654930738867224
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9421.453780355325,
-      "required_pages_per_second_upper_estimate": 0.003758358776270674
+      "required_token_positions_per_second": 9444.93956581112,
+      "required_pages_per_second_upper_estimate": 0.003767727607232775
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18842.90756071065,
-      "required_pages_per_second_upper_estimate": 0.004358975040323701
+      "required_token_positions_per_second": 18889.87913162224,
+      "required_pages_per_second_upper_estimate": 0.004369841086582684
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28264.361341065975,
-      "required_pages_per_second_upper_estimate": 0.0017211513140275433
+      "required_token_positions_per_second": 28334.818697433362,
+      "required_pages_per_second_upper_estimate": 0.0017254417973691374
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28264.361341065975,
-      "required_pages_per_second_upper_estimate": 0.0021758687767950823
+      "required_token_positions_per_second": 28334.818697433362,
+      "required_pages_per_second_upper_estimate": 0.002181292779126689
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28264.361341065975,
-      "required_pages_per_second_upper_estimate": 0.3509487731237316
+      "required_token_positions_per_second": 28334.818697433362,
+      "required_pages_per_second_upper_estimate": 0.3518236176842118
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.5014972019453978,
-    "producer_p95_seconds": 0.3402423820007243,
-    "gpu_update_seconds": 0.9781929853772975,
-    "producer_batches_per_second": 1.9940290715896722,
-    "gpu_batches_per_second": 1.0222931619309163,
-    "producer_headroom": 1.9505452504674223,
+    "producer_batch_seconds": 0.10008276722663823,
+    "producer_p95_seconds": 0.3219244930005516,
+    "gpu_update_seconds": 0.9757606108312394,
+    "producer_batches_per_second": 9.991730122084775,
+    "gpu_batches_per_second": 1.0248415327486025,
+    "producer_headroom": 9.749536687186334,
     "sustainable": true,
-    "buffer_seconds": 60.64796509339244,
+    "buffer_seconds": 60.49715787153684,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252426.17853789695,
+    "generation_bytes_per_second": 250614.6171034554,
     "upload_bytes_per_second": 5181173.629463062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.16478173353845,
+    "estimated_full_shards_per_hour": 53.77606282069918,
     "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": [

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -244,3 +244,4 @@
 {"step": 379400, "loss": 0.5842984765768051, "lr": 0.0003, "grad_norm": 0.1710354123652569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55532897833, "tokens_per_second": 149429.39549429552, "data_wait_s": 1.1393000022508204e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 379500, "loss": 0.5751869510859251, "lr": 0.0003, "grad_norm": 0.15076568877499885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55547540000, "tokens_per_second": 148638.04692342275, "data_wait_s": 1.4183999155648053e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 379600, "loss": 0.5885546300560236, "lr": 0.0003, "grad_norm": 0.16276431622782345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55562066745, "tokens_per_second": 147603.81056144106, "data_wait_s": 1.1120999261038378e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 379700, "loss": 0.5921695046126842, "lr": 0.0003, "grad_norm": 0.15509385439786771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55576775665, "tokens_per_second": 149887.22972496826, "data_wait_s": 9.9280005088076e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c939357999ea496f7ff0e67d59ccccfa2d0c9de12599ffa0182631b21d059e59
-size 524325
+oid sha256:7d6ff3541557bb81a2e5b82f43727fedf6cf5db806a5df0c14c495e9644c036e
+size 526443