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": 76117.85319194965,
-      "required_pages_per_second_upper_estimate": 0.00526600970157166
+      "required_token_positions_per_second": 75232.98061397776,
+      "required_pages_per_second_upper_estimate": 0.00520479216344031
     },
     {
       "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": 9514.731648993706,
-      "required_pages_per_second_upper_estimate": 0.0039836194512430185
+      "required_token_positions_per_second": 9404.12257674722,
+      "required_pages_per_second_upper_estimate": 0.0039373097424734976
     },
     {
       "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": 19029.46329798741,
-      "required_pages_per_second_upper_estimate": 0.003963335364276285
+      "required_token_positions_per_second": 18808.24515349444,
+      "required_pages_per_second_upper_estimate": 0.003917261458693183
     },
     {
       "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": 28544.19494698112,
-      "required_pages_per_second_upper_estimate": 0.0017111809026094562
+      "required_token_positions_per_second": 28212.36773024166,
+      "required_pages_per_second_upper_estimate": 0.0016912883676367485
     },
     {
       "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": 28544.19494698112,
-      "required_pages_per_second_upper_estimate": 0.002219185604483023
+      "required_token_positions_per_second": 28212.36773024166,
+      "required_pages_per_second_upper_estimate": 0.002193387497935206
     },
     {
       "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": 28544.19494698112,
-      "required_pages_per_second_upper_estimate": 0.3661342844111943
+      "required_token_positions_per_second": 28212.36773024166,
+      "required_pages_per_second_upper_estimate": 0.361877961163167
     }
   ],
   "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": 0.11421125123484899,
-    "producer_p95_seconds": 0.3267385139988619,
-    "gpu_update_seconds": 0.9686032502004089,
-    "producer_batches_per_second": 8.755704794300271,
-    "gpu_batches_per_second": 1.0324144584411574,
-    "producer_headroom": 8.480804121554545,
+    "producer_batch_seconds": 0.13801826925754312,
+    "producer_p95_seconds": 0.31696665799972834,
+    "gpu_update_seconds": 0.9799957332316812,
+    "producer_batches_per_second": 7.245417620286141,
+    "gpu_batches_per_second": 1.0204126059838565,
+    "producer_headroom": 7.1004783533620595,
     "sustainable": true,
-    "buffer_seconds": 60.05340151242535,
+    "buffer_seconds": 60.75973546036423,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367914.7681325574,
+    "generation_bytes_per_second": 368703.9001968766,
     "upload_bytes_per_second": 5076634.423330681,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.94594462378065,
+    "estimated_full_shards_per_hour": 79.11527399472926,
     "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

@@ -368,3 +368,4 @@
 {"step": 921800, "loss": 0.5565356612205505, "lr": 3e-05, "grad_norm": 0.3252973719380204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134925109047, "tokens_per_second": 153608.86688052217, "data_wait_s": 1.5084995538927615e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 921900, "loss": 0.525381863117218, "lr": 3e-05, "grad_norm": 0.3327822112093991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134939748858, "tokens_per_second": 152573.92176026368, "data_wait_s": 1.3678996765520424e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922000, "loss": 0.5456743240356445, "lr": 3e-05, "grad_norm": 0.32838502568946504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134954307954, "tokens_per_second": 151984.23384710055, "data_wait_s": 6.5450003603473306e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 922100, "loss": 0.5193803310394287, "lr": 3e-05, "grad_norm": 0.3367932811834201, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134968918631, "tokens_per_second": 153406.63493892195, "data_wait_s": 9.061994205694646e-06, "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 sha256:5aa1a80e0560e97320959a738965b0a0b2616a2461399ecf2a89e1672a749c87
-size 789306
+oid sha256:66617bfd41aea51fac08ff5b9a1e759a040cfafbe5bcf16d540cfa7d32d51c5e
+size 791424