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": 75268.0154420649,
-      "required_pages_per_second_upper_estimate": 0.006524040813477107
+      "required_token_positions_per_second": 75286.12863697638,
+      "required_pages_per_second_upper_estimate": 0.006525610819304561
     },
     {
       "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": 9408.501930258113,
-      "required_pages_per_second_upper_estimate": 0.00375319208961948
+      "required_token_positions_per_second": 9410.766079622048,
+      "required_pages_per_second_upper_estimate": 0.0037540952926528037
     },
     {
       "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": 18817.003860516226,
-      "required_pages_per_second_upper_estimate": 0.004352982675173274
+      "required_token_positions_per_second": 18821.532159244096,
+      "required_pages_per_second_upper_estimate": 0.0043540302173886315
     },
     {
       "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": 28225.50579077434,
-      "required_pages_per_second_upper_estimate": 0.0017187852148741
+      "required_token_positions_per_second": 28232.298238866144,
+      "required_pages_per_second_upper_estimate": 0.00171919883932567
     },
     {
       "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": 28225.50579077434,
-      "required_pages_per_second_upper_estimate": 0.002172877568974585
+      "required_token_positions_per_second": 28232.298238866144,
+      "required_pages_per_second_upper_estimate": 0.0021734004704313916
     },
     {
       "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": 28225.50579077434,
-      "required_pages_per_second_upper_estimate": 0.35046631723027105
+      "required_token_positions_per_second": 28232.298238866144,
+      "required_pages_per_second_upper_estimate": 0.35055065670270985
     }
   ],
   "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": 0.09720022239841342,
-    "producer_p95_seconds": 0.29767800099944,
-    "gpu_update_seconds": 0.9795395768970383,
-    "producer_batches_per_second": 10.288042304071135,
-    "gpu_batches_per_second": 1.0208877962519654,
-    "producer_headroom": 10.077544605628672,
+    "producer_batch_seconds": 0.11328985470311181,
+    "producer_p95_seconds": 0.2914514180010883,
+    "gpu_update_seconds": 0.9793039081011915,
+    "producer_batches_per_second": 8.826915725336633,
+    "gpu_batches_per_second": 1.0211334721812118,
+    "producer_headroom": 8.644233066302029,
     "sustainable": true,
-    "buffer_seconds": 60.73145376761637,
+    "buffer_seconds": 60.716842302273875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488820.12189452705,
-    "upload_bytes_per_second": 4922353.521965594,
+    "generation_bytes_per_second": 485865.229793879,
+    "upload_bytes_per_second": 4959756.806827076,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.88941900851115,
+    "estimated_full_shards_per_hour": 104.25536795007972,
     "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": 0.952104601999963,
-      "bytes": 2531508,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.6972403340005258,
       "bytes": 16066476,
@@ -562,6 +556,12 @@
       "bytes": 11141333,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3496742860006634,
+      "bytes": 7704664,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -219,3 +219,4 @@
 {"step": 216900, "loss": 0.5820107161998749, "lr": 0.0003, "grad_norm": 0.1447592184058356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31747048305, "tokens_per_second": 147250.59297370855, "data_wait_s": 1.1703999916790053e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217000, "loss": 0.5506326481699944, "lr": 0.0003, "grad_norm": 0.13819404198452298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31761729668, "tokens_per_second": 148142.97716309305, "data_wait_s": 1.673400038271211e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217100, "loss": 0.579597033560276, "lr": 0.0003, "grad_norm": 0.15585648008007905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31776458243, "tokens_per_second": 149510.24930505687, "data_wait_s": 1.4774999726796523e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 217200, "loss": 0.5898901894688606, "lr": 0.0003, "grad_norm": 0.15459375768144146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31791071925, "tokens_per_second": 148009.81449124365, "data_wait_s": 1.0289000783814117e-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 sha256:5a5620cc972d84d090d1b98c80004a948ac37df89244475b071bdeb78c5f4bea
-size 471375
+oid sha256:900114f0b17461cf780ebd733895243ffa6defc074fb6052842fd212ec5e5809
+size 473493