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": 76467.86648095111,
-      "required_pages_per_second_upper_estimate": 0.005290224433047503
+      "required_token_positions_per_second": 77026.570005732,
+      "required_pages_per_second_upper_estimate": 0.005328876839262628
     },
     {
       "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": 9558.483310118889,
-      "required_pages_per_second_upper_estimate": 0.004001937358117555
+      "required_token_positions_per_second": 9628.3212507165,
+      "required_pages_per_second_upper_estimate": 0.004031177045463737
     },
     {
       "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": 19116.966620237778,
-      "required_pages_per_second_upper_estimate": 0.003981559998683248
+      "required_token_positions_per_second": 19256.642501433,
+      "required_pages_per_second_upper_estimate": 0.004010650801235521
     },
     {
       "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": 28675.449930356666,
-      "required_pages_per_second_upper_estimate": 0.001719049438448065
+      "required_token_positions_per_second": 28884.963752149502,
+      "required_pages_per_second_upper_estimate": 0.001731609472155463
     },
     {
       "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": 28675.449930356666,
-      "required_pages_per_second_upper_estimate": 0.0022293901021108137
+      "required_token_positions_per_second": 28884.963752149502,
+      "required_pages_per_second_upper_estimate": 0.0022456788802012973
     },
     {
       "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": 28675.449930356666,
-      "required_pages_per_second_upper_estimate": 0.3678178824073148
+      "required_token_positions_per_second": 28884.963752149502,
+      "required_pages_per_second_upper_estimate": 0.3705053007548582
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4396991201953142,
-    "producer_p95_seconds": 0.12538564999999835,
-    "gpu_update_seconds": 0.9641697015094094,
-    "producer_batches_per_second": 2.27428246741954,
-    "gpu_batches_per_second": 1.0371618175042197,
-    "producer_headroom": 2.1927942477599807,
+    "producer_batch_seconds": 0.04085796773437167,
+    "producer_p95_seconds": 0.12610350399995696,
+    "gpu_update_seconds": 0.9571762054900466,
+    "producer_batches_per_second": 24.47503034172579,
+    "gpu_batches_per_second": 1.0447397190447592,
+    "producer_headroom": 23.426916671746852,
     "sustainable": true,
-    "buffer_seconds": 60.74269119509279,
+    "buffer_seconds": 60.30210094587294,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 46,
+    "upload_pages_per_commit": 56,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306376.7640759157,
-    "upload_bytes_per_second": 2943860.1371519086,
+    "generation_bytes_per_second": 308933.42796100024,
+    "upload_bytes_per_second": 3494894.2567762877,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.74132148464301,
+    "estimated_full_shards_per_hour": 66.28992203829294,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 7420176,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0025722640000367,
+      "bytes": 6023360,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -2,3 +2,4 @@
 {"step": 604600, "loss": 0.5752429366111755, "lr": 0.0001, "grad_norm": 0.17567589109079268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88495546086, "tokens_per_second": 149962.14570994265, "data_wait_s": 1.381399999900168e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 604700, "loss": 0.598007082939148, "lr": 0.0001, "grad_norm": 0.15348726900650103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88510111031, "tokens_per_second": 150957.2267887229, "data_wait_s": 1.3324000065040309e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 604800, "loss": 0.551197350025177, "lr": 0.0001, "grad_norm": 0.14659759872408634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88524662114, "tokens_per_second": 138614.38636559332, "data_wait_s": 1.7399000057594094e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 604900, "loss": 0.5269398093223572, "lr": 0.0001, "grad_norm": 0.14970845497430949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88539299377, "tokens_per_second": 151637.69207242082, "data_wait_s": 8.543000035388104e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:315b93e4275424eca44f520438aaa7040581b1499f51b398d5d9eaf1a2c9fa70
-size 8637
+oid sha256:a84ca364d929c7753987215a2d4dcdee6fe48cd467727d8a64dedeeaeb031b6c
+size 10755