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": 76773.49056714427,
-      "required_pages_per_second_upper_estimate": 0.005311368216475927
+      "required_token_positions_per_second": 76786.23254572849,
+      "required_pages_per_second_upper_estimate": 0.005312249736119846
     },
     {
       "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": 9596.686320893034,
-      "required_pages_per_second_upper_estimate": 0.004017932160959107
+      "required_token_positions_per_second": 9598.279068216061,
+      "required_pages_per_second_upper_estimate": 0.004018599011002912
     },
     {
       "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": 19193.372641786067,
-      "required_pages_per_second_upper_estimate": 0.003997473358009468
+      "required_token_positions_per_second": 19196.558136432122,
+      "required_pages_per_second_upper_estimate": 0.003998136812537099
     },
     {
       "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": 28790.0589626791,
-      "required_pages_per_second_upper_estimate": 0.0017259200749379353
+      "required_token_positions_per_second": 28794.83720464818,
+      "required_pages_per_second_upper_estimate": 0.0017262065232480288
     },
     {
       "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": 28790.0589626791,
-      "required_pages_per_second_upper_estimate": 0.002238300450262026
+      "required_token_positions_per_second": 28794.83720464818,
+      "required_pages_per_second_upper_estimate": 0.0022386719375578567
     },
     {
       "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": 28790.0589626791,
-      "required_pages_per_second_upper_estimate": 0.3692879640163556
+      "required_token_positions_per_second": 28794.83720464818,
+      "required_pages_per_second_upper_estimate": 0.3693492541738585
     }
   ],
   "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": 0.05156052077336426,
-    "producer_p95_seconds": 0.16283551800006535,
-    "gpu_update_seconds": 0.960331482330079,
-    "producer_batches_per_second": 19.394683858906866,
-    "gpu_batches_per_second": 1.0413071094719004,
-    "producer_headroom": 18.625325499547287,
+    "producer_batch_seconds": 0.4918614123281486,
+    "producer_p95_seconds": 0.2063163410002744,
+    "gpu_update_seconds": 0.9601721240339899,
+    "producer_batches_per_second": 2.0330930114372205,
+    "gpu_batches_per_second": 1.041479933617194,
+    "producer_headroom": 1.9521192351503367,
     "sustainable": true,
-    "buffer_seconds": 60.50088338679498,
+    "buffer_seconds": 60.490843814141364,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 74,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351231.01162853924,
+    "generation_bytes_per_second": 326108.78536438174,
     "upload_bytes_per_second": 4953259.208074551,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.36599885599263,
+    "estimated_full_shards_per_hour": 69.97535391520108,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -81,3 +81,4 @@
 {"step": 728100, "loss": 0.5516948103904724, "lr": 5e-05, "grad_norm": 0.2244470638871388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106573191688, "tokens_per_second": 149528.11418467318, "data_wait_s": 1.515400072094053e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 728200, "loss": 0.5103027820587158, "lr": 5e-05, "grad_norm": 0.21574655157408928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106587751382, "tokens_per_second": 151112.25882481033, "data_wait_s": 1.3560000297729857e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 728300, "loss": 0.5498045682907104, "lr": 5e-05, "grad_norm": 0.22472514401213922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106602378726, "tokens_per_second": 151615.99751247303, "data_wait_s": 5.373000021791086e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 728400, "loss": 0.5041852593421936, "lr": 5e-05, "grad_norm": 0.22607506738263877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106617058621, "tokens_per_second": 151611.39770952816, "data_wait_s": 1.1276000805082731e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:996fe05973267f6b43fc5c70394331477e90a031962783a0e24a96b9be896b1d
-size 176742
+oid sha256:52ba192d9bfc500205ebed4b1365591f67c1b8fc3dd8a4067ecb8c867885fb99
+size 178860