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": 77184.83700874788,
-      "required_pages_per_second_upper_estimate": 0.005339826117761307
+      "required_token_positions_per_second": 77403.91066291212,
+      "required_pages_per_second_upper_estimate": 0.0053549821414773
     },
     {
       "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": 9648.104626093485,
-      "required_pages_per_second_upper_estimate": 0.004039459931610225
+      "required_token_positions_per_second": 9675.488832864015,
+      "required_pages_per_second_upper_estimate": 0.004050925127137782
     },
     {
       "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": 19296.20925218697,
-      "required_pages_per_second_upper_estimate": 0.004018891511972187
+      "required_token_positions_per_second": 19350.97766572803,
+      "required_pages_per_second_upper_estimate": 0.004030298328172597
     },
     {
       "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": 28944.313878280456,
-      "required_pages_per_second_upper_estimate": 0.0017351674215946183
+      "required_token_positions_per_second": 29026.46649859205,
+      "required_pages_per_second_upper_estimate": 0.0017400923457425103
     },
     {
       "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": 28944.313878280456,
-      "required_pages_per_second_upper_estimate": 0.002250293091454363
+      "required_token_positions_per_second": 29026.46649859205,
+      "required_pages_per_second_upper_estimate": 0.0022566800963324017
     },
     {
       "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": 28944.313878280456,
-      "required_pages_per_second_upper_estimate": 0.3712665804476656
+      "required_token_positions_per_second": 29026.46649859205,
+      "required_pages_per_second_upper_estimate": 0.3723203460524115
     }
   ],
   "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": 0.0869559607108954,
-    "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,
+    "producer_batch_seconds": 0.520029093101499,
+    "producer_p95_seconds": 0.24795101600102498,
+    "gpu_update_seconds": 0.9525100136229495,
+    "producer_batches_per_second": 1.9229693362652318,
+    "gpu_batches_per_second": 1.0498577292604185,
+    "producer_headroom": 1.8316475486825101,
     "sustainable": true,
-    "buffer_seconds": 60.17845188263552,
+    "buffer_seconds": 60.00813085824582,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273038.248349327,
+    "generation_bytes_per_second": 247845.86038035463,
     "upload_bytes_per_second": 5081490.156276682,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.58765209064348,
+    "estimated_full_shards_per_hour": 53.181952081279555,
     "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

@@ -173,3 +173,4 @@
 {"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}
+{"step": 902600, "loss": 0.5620360970497131, "lr": 3e-05, "grad_norm": 0.32152906569240763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132114959058, "tokens_per_second": 151792.44771870453, "data_wait_s": 1.4733002899447456e-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 sha256:070ca30f0496331f4427a7512ffb7b3b42245c2085102fe55527c916571b16f8
-size 373164
+oid sha256:3eb687690fc60cecb590cbf852cf6cd702b22c844e715a62f029edc096fa67d5
+size 375282