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": 75991.0801460436,
-      "required_pages_per_second_upper_estimate": 0.005257239248101887
+      "required_token_positions_per_second": 75760.67868859522,
+      "required_pages_per_second_upper_estimate": 0.005241299540670571
     },
     {
       "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": 9498.88501825545,
-      "required_pages_per_second_upper_estimate": 0.003976984797868191
+      "required_token_positions_per_second": 9470.084836074402,
+      "required_pages_per_second_upper_estimate": 0.003964926763005179
     },
     {
       "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": 18997.7700365109,
-      "required_pages_per_second_upper_estimate": 0.003956734493718238
+      "required_token_positions_per_second": 18940.169672148804,
+      "required_pages_per_second_upper_estimate": 0.003944737856845372
     },
     {
       "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": 28496.65505476635,
-      "required_pages_per_second_upper_estimate": 0.0017083309586603931
+      "required_token_positions_per_second": 28410.254508223206,
+      "required_pages_per_second_upper_estimate": 0.0017031513778211277
     },
     {
       "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": 28496.65505476635,
-      "required_pages_per_second_upper_estimate": 0.0022154895869692113
+      "required_token_positions_per_second": 28410.254508223206,
+      "required_pages_per_second_upper_estimate": 0.0022087723245113227
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28496.65505476635,
-      "required_pages_per_second_upper_estimate": 0.3655244937182226
+      "required_token_positions_per_second": 28410.254508223206,
+      "required_pages_per_second_upper_estimate": 0.3644162402768462
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2681699248831251,
-    "producer_p95_seconds": 0.3320305599991116,
-    "gpu_update_seconds": 0.9702191343813735,
-    "producer_batches_per_second": 3.728978931682306,
-    "gpu_batches_per_second": 1.0306949889600097,
-    "producer_headroom": 3.617926711223186,
+    "producer_batch_seconds": 0.20588724810943404,
+    "producer_p95_seconds": 0.31248633100040024,
+    "gpu_update_seconds": 0.9731697402428205,
+    "producer_batches_per_second": 4.857027373878327,
+    "gpu_batches_per_second": 1.027569969192101,
+    "producer_headroom": 4.726712067789441,
     "sustainable": true,
-    "buffer_seconds": 60.15358633164516,
+    "buffer_seconds": 60.336523895054874,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293934.8236481574,
+    "generation_bytes_per_second": 293941.9438503281,
     "upload_bytes_per_second": 4903113.124838578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.07157070239584,
+    "estimated_full_shards_per_hour": 63.07309853203184,
     "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": [

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 927400, "loss": 0.5002279281616211, "lr": 3e-05, "grad_norm": 0.32860060915400224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135745010409, "tokens_per_second": 153240.5829542607, "data_wait_s": 1.6832003893796355e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927500, "loss": 0.523468554019928, "lr": 3e-05, "grad_norm": 0.34131293320080286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135759655387, "tokens_per_second": 152516.0676678596, "data_wait_s": 8.820999937597662e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 927600, "loss": 0.5498864650726318, "lr": 3e-05, "grad_norm": 0.3313419076080202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135774307827, "tokens_per_second": 153132.47373963153, "data_wait_s": 1.359300222247839e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 927700, "loss": 0.549930989742279, "lr": 3e-05, "grad_norm": 0.33515906406952983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135788903980, "tokens_per_second": 153009.45018511394, "data_wait_s": 1.1962998542003334e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a669ce431b9d8137b7c9d73f4222fb1ccb0bf6d9c551616b00102da5a56d716
-size 908697
+oid sha256:6f40b6d0904521e50915dc52bfb5bf00189fa95cc0e7b1d90ef2cf10ec1b6f93
+size 910815