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": 75678.43738681504,
-      "required_pages_per_second_upper_estimate": 0.00655961514744314
+      "required_token_positions_per_second": 75336.96992088645,
+      "required_pages_per_second_upper_estimate": 0.00653001761293783
     },
     {
       "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": 9459.80467335188,
-      "required_pages_per_second_upper_estimate": 0.0037736575208839477
+      "required_token_positions_per_second": 9417.121240110806,
+      "required_pages_per_second_upper_estimate": 0.0037566304611898857
     },
     {
       "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": 18919.60934670376,
-      "required_pages_per_second_upper_estimate": 0.004376718648607879
+      "required_token_positions_per_second": 18834.242480221612,
+      "required_pages_per_second_upper_estimate": 0.00435697052645277
     },
     {
       "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": 28379.414020055643,
-      "required_pages_per_second_upper_estimate": 0.0017281574185432588
+      "required_token_positions_per_second": 28251.363720332418,
+      "required_pages_per_second_upper_estimate": 0.0017203598271181154
     },
     {
       "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": 28379.414020055643,
-      "required_pages_per_second_upper_estimate": 0.002184725850509906
+      "required_token_positions_per_second": 28251.363720332418,
+      "required_pages_per_second_upper_estimate": 0.002174868183971296
     },
     {
       "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": 28379.414020055643,
-      "required_pages_per_second_upper_estimate": 0.3523773423402367
+      "required_token_positions_per_second": 28251.363720332418,
+      "required_pages_per_second_upper_estimate": 0.35078738617445915
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.8217937024063247,
-    "producer_p95_seconds": 0.3307132059999276,
-    "gpu_update_seconds": 0.974227303652614,
-    "producer_batches_per_second": 1.2168504054872444,
-    "gpu_batches_per_second": 1.0264545001466885,
-    "producer_headroom": 1.1854888894864282,
+    "producer_batch_seconds": 0.10192410661721851,
+    "producer_p95_seconds": 0.2983065189982881,
+    "gpu_update_seconds": 0.9786430231720751,
+    "producer_batches_per_second": 9.811221635285497,
+    "gpu_batches_per_second": 1.021823051227301,
+    "producer_headroom": 9.60168360216707,
     "sustainable": true,
-    "buffer_seconds": 60.40209282646207,
+    "buffer_seconds": 60.67586743666865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252939.77857298334,
-    "upload_bytes_per_second": 2830509.1987755364,
+    "generation_bytes_per_second": 251499.57862477956,
+    "upload_bytes_per_second": 2830244.747027332,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.27498834507108,
+    "estimated_full_shards_per_hour": 53.965954962325476,
     "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": 1.0575754780002171,
-      "bytes": 10536411,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 3.8299359349985025,
       "bytes": 2330234,
@@ -562,6 +556,12 @@
       "bytes": 6519513,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.2462770620004449,
+      "bytes": 11030793,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -286,3 +286,4 @@
 {"step": 223600, "loss": 0.5933656673878431, "lr": 0.0003, "grad_norm": 0.1388028229036762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32728172930, "tokens_per_second": 150344.67360773784, "data_wait_s": 1.0833999112946913e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223700, "loss": 0.6178768519312143, "lr": 0.0003, "grad_norm": 0.14937646539716934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32742841280, "tokens_per_second": 106503.98873140896, "data_wait_s": 1.2504999176599085e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223800, "loss": 0.5782792679965496, "lr": 0.0003, "grad_norm": 0.1459301303786713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32757501567, "tokens_per_second": 149032.4429403084, "data_wait_s": 1.0058000043500215e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 223900, "loss": 0.5707023125141859, "lr": 0.0003, "grad_norm": 0.15491452613289144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32772181501, "tokens_per_second": 148682.36673024838, "data_wait_s": 1.8599999748403206e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03810318754b06712f2349a85688ff00261e568bad110dc586a4d44916ae828f
-size 614064
+oid sha256:bcf4f8e27aa07f5380a68064a12de6a3170f59ad071d39dbf9cc09f10db8a760
+size 616182