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": 76525.33649736669,
-      "required_pages_per_second_upper_estimate": 0.005294200342131942
+      "required_token_positions_per_second": 76818.80891629407,
+      "required_pages_per_second_upper_estimate": 0.0053145034450231676
     },
     {
       "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": 9565.667062170836,
-      "required_pages_per_second_upper_estimate": 0.004004945045087904
+      "required_token_positions_per_second": 9602.35111453676,
+      "required_pages_per_second_upper_estimate": 0.0040203038917633925
     },
     {
       "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": 19131.334124341673,
-      "required_pages_per_second_upper_estimate": 0.003984552370891526
+      "required_token_positions_per_second": 19204.70222907352,
+      "required_pages_per_second_upper_estimate": 0.003999833012260135
     },
     {
       "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": 28697.001186512513,
-      "required_pages_per_second_upper_estimate": 0.0017203414033477452
+      "required_token_positions_per_second": 28807.053343610278,
+      "required_pages_per_second_upper_estimate": 0.0017269388621258422
     },
     {
       "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": 28697.001186512513,
-      "required_pages_per_second_upper_estimate": 0.002231065617482973
+      "required_token_positions_per_second": 28807.053343610278,
+      "required_pages_per_second_upper_estimate": 0.0022396216886290433
     },
     {
       "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": 28697.001186512513,
-      "required_pages_per_second_upper_estimate": 0.36809431878134596
+      "required_token_positions_per_second": 28807.053343610278,
+      "required_pages_per_second_upper_estimate": 0.36950594968779615
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5077920868516514,
-    "producer_p95_seconds": 0.2842293260000588,
-    "gpu_update_seconds": 0.9634456165055484,
-    "producer_batches_per_second": 1.9693099319449305,
-    "gpu_batches_per_second": 1.037941304489023,
-    "producer_headroom": 1.897323021473183,
+    "producer_batch_seconds": 0.08348779447652532,
+    "producer_p95_seconds": 0.25826175900147064,
+    "gpu_update_seconds": 0.959764946112846,
+    "producer_batches_per_second": 11.977798746152947,
+    "gpu_batches_per_second": 1.0419217789210893,
+    "producer_headroom": 11.495871368151999,
     "sustainable": true,
-    "buffer_seconds": 60.69707383984955,
+    "buffer_seconds": 60.4651916051093,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265106.2507338192,
-    "upload_bytes_per_second": 3885076.6956475764,
+    "generation_bytes_per_second": 253379.39607922127,
+    "upload_bytes_per_second": 3918962.455501599,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.88563004980976,
+    "estimated_full_shards_per_hour": 54.369320028137956,
     "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": [
-    {
-      "seconds": 1.6447518009990745,
-      "bytes": 6129393,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.743864690999544,
       "bytes": 1054003,
@@ -562,6 +556,12 @@
       "bytes": 2323345,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.9117556889978005,
+      "bytes": 10784078,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 820700, "loss": 0.5129390954971313, "lr": 3e-05, "grad_norm": 0.2830667316656344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120127553733, "tokens_per_second": 149909.6327589666, "data_wait_s": 1.0944000678136945e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 820800, "loss": 0.48247233033180237, "lr": 3e-05, "grad_norm": 0.265050563943269, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120142253156, "tokens_per_second": 152112.31392620542, "data_wait_s": 1.0331998055335134e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 820900, "loss": 0.5293641090393066, "lr": 3e-05, "grad_norm": 0.2794738462263193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120156806130, "tokens_per_second": 150516.99113490837, "data_wait_s": 2.1830001060152426e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 821000, "loss": 0.5010907053947449, "lr": 3e-05, "grad_norm": 0.2722542619171817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120171463232, "tokens_per_second": 151998.30985808626, "data_wait_s": 1.1166997865075245e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:033d54fb8b98eb8f2e111e94138a6edf1754c68d22916a575611b357a5f4e6ca
-size 445959
+oid sha256:7144f1f0c9f7668307a766a1074d1df105cddeceb6d0359f7581a7d1297a0692
+size 448077