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": 76520.83730199195,
-      "required_pages_per_second_upper_estimate": 0.0052938890773563466
+      "required_token_positions_per_second": 76683.26448242246,
+      "required_pages_per_second_upper_estimate": 0.005305126166581512
     },
     {
       "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": 9565.104662748994,
-      "required_pages_per_second_upper_estimate": 0.004004709580194214
+      "required_token_positions_per_second": 9585.408060302807,
+      "required_pages_per_second_upper_estimate": 0.004013210188766842
     },
     {
       "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": 19130.209325497988,
-      "required_pages_per_second_upper_estimate": 0.003984318104955327
+      "required_token_positions_per_second": 19170.816120605614,
+      "required_pages_per_second_upper_estimate": 0.003992775429503046
     },
     {
       "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": 28695.31398824698,
-      "required_pages_per_second_upper_estimate": 0.0017202402583879292
+      "required_token_positions_per_second": 28756.224180908423,
+      "required_pages_per_second_upper_estimate": 0.0017238917314335036
     },
     {
       "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": 28695.31398824698,
-      "required_pages_per_second_upper_estimate": 0.0022309344452390288
+      "required_token_positions_per_second": 28756.224180908423,
+      "required_pages_per_second_upper_estimate": 0.0022356699448028328
     },
     {
       "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": 28695.31398824698,
-      "required_pages_per_second_upper_estimate": 0.36807267721356807
+      "required_token_positions_per_second": 28756.224180908423,
+      "required_pages_per_second_upper_estimate": 0.3688539677647596
     }
   ],
   "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": 0.13519453901579936,
-    "producer_p95_seconds": 0.2767468899983214,
-    "gpu_update_seconds": 0.963502264213734,
-    "producer_batches_per_second": 7.396748472829484,
-    "gpu_batches_per_second": 1.0378802802462017,
-    "producer_headroom": 7.126783901390686,
+    "producer_batch_seconds": 0.08606309461700334,
+    "producer_p95_seconds": 0.30477311000140617,
+    "gpu_update_seconds": 0.9614614153118133,
+    "producer_batches_per_second": 11.619382320031422,
+    "gpu_batches_per_second": 1.0400833398766067,
+    "producer_headroom": 11.17158777046647,
     "sustainable": true,
-    "buffer_seconds": 60.70064264546524,
+    "buffer_seconds": 60.57206916464423,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307928.9779755035,
-    "upload_bytes_per_second": 4906086.108551338,
+    "generation_bytes_per_second": 307609.7138713683,
+    "upload_bytes_per_second": 4810650.709177618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.07439045380428,
+    "estimated_full_shards_per_hour": 66.00588380914485,
     "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": 1.4002689310000278,
-      "bytes": 11880783,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.1156020849994093,
       "bytes": 6256874,
@@ -562,6 +556,12 @@
       "bytes": 11971576,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 3.1373434490014915,
+      "bytes": 11875917,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -275,3 +275,4 @@
 {"step": 870000, "loss": 0.4904373586177826, "lr": 3e-05, "grad_norm": 0.3192784582104852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127343745288, "tokens_per_second": 150244.0324024395, "data_wait_s": 9.524999768473208e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 870100, "loss": 0.5325773358345032, "lr": 3e-05, "grad_norm": 0.32444056901834195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127358306698, "tokens_per_second": 141699.04062335717, "data_wait_s": 1.3484001101460308e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 870200, "loss": 0.5269530415534973, "lr": 3e-05, "grad_norm": 0.30391292623804333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127372861825, "tokens_per_second": 151274.96112900405, "data_wait_s": 1.0161998943658546e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 870300, "loss": 0.5636822581291199, "lr": 3e-05, "grad_norm": 0.2987178258372232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127387521226, "tokens_per_second": 152497.3983713821, "data_wait_s": 1.0017000022344291e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afdc6f8aa2949c1d5814aced966903bbb461e07739948f6dc8e1c4c08163a463
-size 591549
+oid sha256:ddc360f40f4209e6122f61c60fcd525d22864296499a581fffec585b0a8b163e
+size 593667