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": 77008.83232556813,
-      "required_pages_per_second_upper_estimate": 0.0053276497053918025
+      "required_token_positions_per_second": 77184.83700874788,
+      "required_pages_per_second_upper_estimate": 0.005339826117761307
     },
     {
       "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": 9626.104040696016,
-      "required_pages_per_second_upper_estimate": 0.004030248745928771
+      "required_token_positions_per_second": 9648.104626093485,
+      "required_pages_per_second_upper_estimate": 0.004039459931610225
     },
     {
       "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": 19252.208081392033,
-      "required_pages_per_second_upper_estimate": 0.004009727228484508
+      "required_token_positions_per_second": 19296.20925218697,
+      "required_pages_per_second_upper_estimate": 0.004018891511972187
     },
     {
       "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": 28878.31212208805,
-      "required_pages_per_second_upper_estimate": 0.00173121071708973
+      "required_token_positions_per_second": 28944.313878280456,
+      "required_pages_per_second_upper_estimate": 0.0017351674215946183
     },
     {
       "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": 28878.31212208805,
-      "required_pages_per_second_upper_estimate": 0.002245161745221448
+      "required_token_positions_per_second": 28944.313878280456,
+      "required_pages_per_second_upper_estimate": 0.002250293091454363
     },
     {
       "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": 28878.31212208805,
-      "required_pages_per_second_upper_estimate": 0.3704199807863938
+      "required_token_positions_per_second": 28944.313878280456,
+      "required_pages_per_second_upper_estimate": 0.3712665804476656
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08725946541417784,
-    "producer_p95_seconds": 0.2630753070006904,
-    "gpu_update_seconds": 0.9573966748164957,
-    "producer_batches_per_second": 11.460074792499483,
-    "gpu_batches_per_second": 1.044499136360245,
-    "producer_headroom": 10.971837499487348,
+    "producer_batch_seconds": 0.0869559607108954,
+    "producer_p95_seconds": 0.25260094799887156,
+    "gpu_update_seconds": 0.9552135219465956,
+    "producer_batches_per_second": 11.500074196462787,
+    "gpu_batches_per_second": 1.0468863526577132,
+    "producer_headroom": 10.985026375850383,
     "sustainable": true,
-    "buffer_seconds": 60.31599051343923,
+    "buffer_seconds": 60.17845188263552,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286040.64531836205,
-    "upload_bytes_per_second": 5042307.947079274,
+    "generation_bytes_per_second": 273038.248349327,
+    "upload_bytes_per_second": 5081490.156276682,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.3776637998881,
+    "estimated_full_shards_per_hour": 58.58765209064348,
     "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": 1.1927761070000997,
-      "bytes": 2669572,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1629383399995277,
       "bytes": 10339749,
@@ -562,6 +556,12 @@
       "bytes": 8230136,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2011196589992323,
+      "bytes": 6075797,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -172,3 +172,4 @@
 {"step": 902200, "loss": 0.5325385332107544, "lr": 3e-05, "grad_norm": 0.3249382554368432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132056366016, "tokens_per_second": 153416.69024096013, "data_wait_s": 1.0091000149259344e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 902300, "loss": 0.5309786200523376, "lr": 3e-05, "grad_norm": 0.3312332507631469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132070973656, "tokens_per_second": 150914.83163535254, "data_wait_s": 1.4658999134553596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 902400, "loss": 0.5159944295883179, "lr": 3e-05, "grad_norm": 0.3183463962950348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132085600369, "tokens_per_second": 152085.38632008614, "data_wait_s": 1.704399983282201e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 902500, "loss": 0.5932426452636719, "lr": 3e-05, "grad_norm": 0.3489375657512693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132100369949, "tokens_per_second": 138983.25193942292, "data_wait_s": 1.4033998013474047e-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 sha256:9ca3d58f38144cfeae75cd4efecc3193f02803932f7ef54a9f0f295f51c9174f
-size 371046
+oid sha256:070ca30f0496331f4427a7512ffb7b3b42245c2085102fe55527c916571b16f8
+size 373164