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": 75515.5335159033,
-      "required_pages_per_second_upper_estimate": 0.005224339828808576
+      "required_token_positions_per_second": 76005.46426447974,
+      "required_pages_per_second_upper_estimate": 0.005258234374791051
     },
     {
       "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": 9439.441689487912,
-      "required_pages_per_second_upper_estimate": 0.003952097117431082
+      "required_token_positions_per_second": 9500.683033059968,
+      "required_pages_per_second_upper_estimate": 0.003977737589120024
     },
     {
       "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": 18878.883378975825,
-      "required_pages_per_second_upper_estimate": 0.003931973538205678
+      "required_token_positions_per_second": 19001.366066119936,
+      "required_pages_per_second_upper_estimate": 0.003957483451852122
     },
     {
       "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": 28318.32506846374,
-      "required_pages_per_second_upper_estimate": 0.0016976403482756744
+      "required_token_positions_per_second": 28502.0490991799,
+      "required_pages_per_second_upper_estimate": 0.0017086543233867568
     },
     {
       "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": 28318.32506846374,
-      "required_pages_per_second_upper_estimate": 0.002201625214924896
+      "required_token_positions_per_second": 28502.0490991799,
+      "required_pages_per_second_upper_estimate": 0.002215908950196471
     },
     {
       "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": 28318.32506846374,
-      "required_pages_per_second_upper_estimate": 0.3632370681297538
+      "required_token_positions_per_second": 28502.0490991799,
+      "required_pages_per_second_upper_estimate": 0.36559368272828596
     }
   ],
   "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": 0.07133185918744545,
-    "producer_p95_seconds": 0.1621071909994498,
-    "gpu_update_seconds": 0.9763289295237932,
-    "producer_batches_per_second": 14.018981299396751,
-    "gpu_batches_per_second": 1.0242449749878377,
-    "producer_headroom": 13.687137005054106,
+    "producer_batch_seconds": 0.04738611721879238,
+    "producer_p95_seconds": 0.16571094800019637,
+    "gpu_update_seconds": 0.9700355193337845,
+    "producer_batches_per_second": 21.10322724655356,
+    "gpu_batches_per_second": 1.0308900860525139,
+    "producer_headroom": 20.47088000172945,
     "sustainable": true,
-    "buffer_seconds": 60.532393630475184,
+    "buffer_seconds": 60.14220219869464,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 205841.18384926824,
-    "upload_bytes_per_second": 3210988.511302942,
+    "generation_bytes_per_second": 208706.7415540613,
+    "upload_bytes_per_second": 3142118.2507490185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.16872631653343,
+    "estimated_full_shards_per_hour": 44.783608293212694,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 12327681,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 2.076565071999539,
+      "bytes": 6113882,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 724400, "loss": 0.4595949649810791, "lr": 5e-05, "grad_norm": 0.21529503899523042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106032156398, "tokens_per_second": 115167.31695636589, "data_wait_s": 1.4557999747921713e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724500, "loss": 0.562905490398407, "lr": 5e-05, "grad_norm": 0.20877419289320268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106046662623, "tokens_per_second": 148829.5518433356, "data_wait_s": 8.355000318260863e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 724600, "loss": 0.50150066614151, "lr": 5e-05, "grad_norm": 0.2281625472419726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106061257752, "tokens_per_second": 150549.5300922586, "data_wait_s": 1.464100023440551e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 724700, "loss": 0.5295780301094055, "lr": 5e-05, "grad_norm": 0.21179198168285204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106076008572, "tokens_per_second": 152179.90967603103, "data_wait_s": 1.2849999620812014e-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 sha256:0a957561d0d8b6c1e1dafc4a1a5af40cc646e8358e3146ef109d855e0728ba4b
-size 97593
+oid sha256:674f36e5e17f6b7048099f51cf547bbe0fe33aa57205fec3e7233f249bcbaaf4
+size 99711