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": 75099.02718546317,
-      "required_pages_per_second_upper_estimate": 0.005195524954440894
+      "required_token_positions_per_second": 76932.506859015,
+      "required_pages_per_second_upper_estimate": 0.005322369332516175
     },
     {
       "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": 9387.378398182896,
-      "required_pages_per_second_upper_estimate": 0.003930299304566825
+      "required_token_positions_per_second": 9616.563357376876,
+      "required_pages_per_second_upper_estimate": 0.00402625426105517
     },
     {
       "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": 18774.756796365793,
-      "required_pages_per_second_upper_estimate": 0.003910286717050652
+      "required_token_positions_per_second": 19233.12671475375,
+      "required_pages_per_second_upper_estimate": 0.0040057530830232314
     },
     {
       "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": 28162.135194548693,
-      "required_pages_per_second_upper_estimate": 0.001688277003822596
+      "required_token_positions_per_second": 28849.690072130627,
+      "required_pages_per_second_upper_estimate": 0.0017294948688981223
     },
     {
       "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": 28162.135194548693,
-      "required_pages_per_second_upper_estimate": 0.0021894821392346516
+      "required_token_positions_per_second": 28849.690072130627,
+      "required_pages_per_second_upper_estimate": 0.0022429365067323515
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28162.135194548693,
-      "required_pages_per_second_upper_estimate": 0.36123363213079224
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28849.690072130627,
+      "required_pages_per_second_upper_estimate": 0.3700528478615029
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12041136084366144,
-    "producer_p95_seconds": 0.3402846850003698,
-    "gpu_update_seconds": 0.9817437424045812,
-    "producer_batches_per_second": 8.304864200466685,
-    "gpu_batches_per_second": 1.0185957463306095,
-    "producer_headroom": 8.153248460327994,
+    "producer_batch_seconds": 0.09622676377330208,
+    "producer_p95_seconds": 0.34386066899605794,
+    "gpu_update_seconds": 0.9583465170985846,
+    "producer_batches_per_second": 10.392119206626047,
+    "gpu_batches_per_second": 1.0434639059653728,
+    "producer_headroom": 9.959251246943378,
     "sustainable": true,
-    "buffer_seconds": 60.86811202908403,
+    "buffer_seconds": 60.375830577210834,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280528.61719710485,
-    "upload_bytes_per_second": 3876412.9613341535,
+    "generation_bytes_per_second": 280682.7380025995,
+    "upload_bytes_per_second": 3845643.949886112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.1949108785139,
+    "estimated_full_shards_per_hour": 60.22798161562432,
     "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": 1.4348496300008264,
-      "bytes": 6710244,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.629695798998,
       "bytes": 10655327,
@@ -562,6 +556,12 @@
       "bytes": 886900,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 2.2302413529978367,
+      "bytes": 6385571,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -367,3 +367,4 @@
 {"step": 879200, "loss": 0.5450117588043213, "lr": 3e-05, "grad_norm": 0.3068843426753608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128690530526, "tokens_per_second": 153357.47921515413, "data_wait_s": 8.413000614382327e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 879300, "loss": 0.545331597328186, "lr": 3e-05, "grad_norm": 0.31861703193024676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128705281439, "tokens_per_second": 152969.91907331857, "data_wait_s": 9.764000424183905e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 879400, "loss": 0.5143944621086121, "lr": 3e-05, "grad_norm": 0.29641323101373607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128719770269, "tokens_per_second": 150172.49895940014, "data_wait_s": 1.1741998605430126e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 879500, "loss": 0.5753275752067566, "lr": 3e-05, "grad_norm": 0.3235909014660641, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128734278305, "tokens_per_second": 132483.3379214986, "data_wait_s": 1.0771000233944505e-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 sha256:545c6a418fc4e80a3ff01f20753d2f815c0e32cf5790ae921b95289bcfac38c4
-size 787188
+oid sha256:22adca7dcf336424d690ba714f050c3069309ac68d8939d618b83b3da480ee26
+size 789306