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": 75244.64309426477,
-      "required_pages_per_second_upper_estimate": 0.005205599000887248
+      "required_token_positions_per_second": 75670.08511586877,
+      "required_pages_per_second_upper_estimate": 0.005235032067103307
     },
     {
       "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": 9405.580386783096,
-      "required_pages_per_second_upper_estimate": 0.003937920097092984
+      "required_token_positions_per_second": 9458.760639483597,
+      "required_pages_per_second_upper_estimate": 0.003960185558368724
     },
     {
       "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": 18811.160773566193,
-      "required_pages_per_second_upper_estimate": 0.0039178687054640575
+      "required_token_positions_per_second": 18917.521278967193,
+      "required_pages_per_second_upper_estimate": 0.003940020793823936
     },
     {
       "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": 28216.74116034929,
-      "required_pages_per_second_upper_estimate": 0.0016915505480938598
+      "required_token_positions_per_second": 28376.281918450793,
+      "required_pages_per_second_upper_estimate": 0.0017011147729374125
     },
     {
       "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": 28216.74116034929,
-      "required_pages_per_second_upper_estimate": 0.002193727512889388
+      "required_token_positions_per_second": 28376.281918450793,
+      "required_pages_per_second_upper_estimate": 0.0022061311050861477
     },
     {
       "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": 28216.74116034929,
-      "required_pages_per_second_upper_estimate": 0.3619340588287643
+      "required_token_positions_per_second": 28376.281918450793,
+      "required_pages_per_second_upper_estimate": 0.3639804763721706
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11825147675756398,
-    "producer_p95_seconds": 0.35935600099764997,
-    "gpu_update_seconds": 0.9798438396157351,
-    "producer_batches_per_second": 8.45655401031628,
-    "gpu_batches_per_second": 1.0205707884964297,
-    "producer_headroom": 8.286102351386146,
+    "producer_batch_seconds": 0.09808939145329987,
+    "producer_p95_seconds": 0.356508610995661,
+    "gpu_update_seconds": 0.9743348363769515,
+    "producer_batches_per_second": 10.194782383537344,
+    "gpu_batches_per_second": 1.0263412152217446,
+    "producer_headroom": 9.933131625562485,
     "sustainable": true,
-    "buffer_seconds": 60.75031805617557,
+    "buffer_seconds": 60.408759855370995,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483596.7439536269,
+    "generation_bytes_per_second": 504219.6339822685,
     "upload_bytes_per_second": 5554724.018092232,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.76860369640927,
+    "estimated_full_shards_per_hour": 108.19379582024614,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 599100, "loss": 0.5877408385276794, "lr": 0.0003, "grad_norm": 0.1686098600321335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87690455460, "tokens_per_second": 152727.06789891224, "data_wait_s": 1.0186995496042073e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599200, "loss": 0.6008965969085693, "lr": 0.0003, "grad_norm": 0.19879451628919603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87705151674, "tokens_per_second": 153247.47532828877, "data_wait_s": 1.5926998457871377e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599300, "loss": 0.573486328125, "lr": 0.0003, "grad_norm": 0.18624040246734963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87719923972, "tokens_per_second": 153211.06162361678, "data_wait_s": 1.0664000001270324e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 599400, "loss": 0.582827627658844, "lr": 0.0003, "grad_norm": 0.1467533053899549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87734562111, "tokens_per_second": 151372.81203418892, "data_wait_s": 1.3983997632749379e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58526af0b44193e6e6b5a399ed63149267e7515746155af8fdad069d2bdb90f5
-size 900225
+oid sha256:fce46646d096fa0d307b165759427f84c16e552e1a31316b14e0f9f679b40135
+size 902343