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": 77041.47555260362,
-      "required_pages_per_second_upper_estimate": 0.005329908039580839
+      "required_token_positions_per_second": 76584.59811982358,
+      "required_pages_per_second_upper_estimate": 0.005298300198679419
     },
     {
       "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": 9630.184444075452,
-      "required_pages_per_second_upper_estimate": 0.004031957125615229
+      "required_token_positions_per_second": 9573.074764977948,
+      "required_pages_per_second_upper_estimate": 0.004008046495562817
     },
     {
       "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": 19260.368888150904,
-      "required_pages_per_second_upper_estimate": 0.004011426909317434
+      "required_token_positions_per_second": 19146.149529955896,
+      "required_pages_per_second_upper_estimate": 0.003987638029172449
     },
     {
       "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": 28890.553332226358,
-      "required_pages_per_second_upper_estimate": 0.001731944558946276
+      "required_token_positions_per_second": 28719.224294933843,
+      "required_pages_per_second_upper_estimate": 0.0017216736447648823
     },
     {
       "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": 28890.553332226358,
-      "required_pages_per_second_upper_estimate": 0.0022461134454663085
+      "required_token_positions_per_second": 28719.224294933843,
+      "required_pages_per_second_upper_estimate": 0.002232793366413608
     },
     {
       "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": 28890.553332226358,
-      "required_pages_per_second_upper_estimate": 0.37057699788646065
+      "required_token_positions_per_second": 28719.224294933843,
+      "required_pages_per_second_upper_estimate": 0.36837937295486006
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08469926949194928,
-    "producer_p95_seconds": 0.22858502099916223,
-    "gpu_update_seconds": 0.9569910164773364,
-    "producer_batches_per_second": 11.806477269500544,
-    "gpu_batches_per_second": 1.0449418884630481,
-    "producer_headroom": 11.298692683155892,
+    "producer_batch_seconds": 0.1855617822656086,
+    "producer_p95_seconds": 0.21410644500065246,
+    "gpu_update_seconds": 0.9627000964951964,
+    "producer_batches_per_second": 5.38904071619998,
+    "gpu_batches_per_second": 1.0387450916859753,
+    "producer_headroom": 5.188030017502262,
     "sustainable": true,
-    "buffer_seconds": 60.29043403807219,
+    "buffer_seconds": 60.65010607919737,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467193.4694828884,
-    "upload_bytes_per_second": 5242620.329022737,
+    "generation_bytes_per_second": 489947.4378135393,
+    "upload_bytes_per_second": 5220228.23420777,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.2488428436755,
+    "estimated_full_shards_per_hour": 105.13131476215968,
     "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": 1.055939247000424,
-      "bytes": 7659150,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1426443119999021,
       "bytes": 11681956,
@@ -562,6 +556,12 @@
       "bytes": 2343317,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1246145429977332,
+      "bytes": 6175304,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -210,3 +210,4 @@
 {"step": 578100, "loss": 0.5821698307991028, "lr": 0.0003, "grad_norm": 0.16252691830957244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84616504976, "tokens_per_second": 152554.61799714432, "data_wait_s": 1.166199945146218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 578200, "loss": 0.567690908908844, "lr": 0.0003, "grad_norm": 0.1695562484367585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84631326615, "tokens_per_second": 154239.39607172104, "data_wait_s": 9.08000220078975e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 578300, "loss": 0.564687967300415, "lr": 0.0003, "grad_norm": 0.15118391940111528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84646007738, "tokens_per_second": 153030.38577464016, "data_wait_s": 1.004899968393147e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 578400, "loss": 0.5859547853469849, "lr": 0.0003, "grad_norm": 0.16622298513679798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84660777057, "tokens_per_second": 154022.36304698404, "data_wait_s": 7.766000635456294e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1493a7d801fe75197145f79f168c4aff6f4416a868d91524925df15e24379e7
-size 452313
+oid sha256:581ad47cf291839d99f5f33f986194deefd6b7fe32f00a7d9e04cb1ca181110d
+size 454431