Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74974.8518222558,
-      "required_pages_per_second_upper_estimate": 0.006498630133928455
+      "required_token_positions_per_second": 75726.17348104177,
+      "required_pages_per_second_upper_estimate": 0.006563752791104688
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9371.856477781976,
-      "required_pages_per_second_upper_estimate": 0.003738573670728409
+      "required_token_positions_per_second": 9465.771685130221,
+      "required_pages_per_second_upper_estimate": 0.003776037851097104
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18743.71295556395,
-      "required_pages_per_second_upper_estimate": 0.004336028114188438
+      "required_token_positions_per_second": 18931.543370260442,
+      "required_pages_per_second_upper_estimate": 0.004379479374926076
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28115.56943334593,
-      "required_pages_per_second_upper_estimate": 0.0017120906674982005
+      "required_token_positions_per_second": 28397.315055390667,
+      "required_pages_per_second_upper_estimate": 0.001729247497675688
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28115.56943334593,
-      "required_pages_per_second_upper_estimate": 0.002164414363856431
+      "required_token_positions_per_second": 28397.315055390667,
+      "required_pages_per_second_upper_estimate": 0.002186103921763234
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28115.56943334593,
-      "required_pages_per_second_upper_estimate": 0.34910127560433
+      "required_token_positions_per_second": 28397.315055390667,
+      "required_pages_per_second_upper_estimate": 0.3525996132881864
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.06493126138281546,
+    "producer_batch_seconds": 0.0658548407656454,
     "producer_p95_seconds": 0.14561187499975858,
-    "gpu_update_seconds": 0.983369732757702,
-    "producer_batches_per_second": 15.40090210329192,
-    "gpu_batches_per_second": 1.0169115101759956,
-    "producer_headroom": 15.144780985541704,
+    "gpu_update_seconds": 0.9736131724450329,
+    "producer_batches_per_second": 15.184912580058526,
+    "gpu_batches_per_second": 1.0271019623622202,
+    "producer_headroom": 14.78423091037127,
     "sustainable": true,
-    "buffer_seconds": 60.968923430977526,
+    "buffer_seconds": 60.364016691592035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 71,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326246.4619777408,
-    "upload_bytes_per_second": 4952983.905081139,
+    "generation_bytes_per_second": 326268.46645770926,
+    "upload_bytes_per_second": 5015459.328009876,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.0048961114804,
+    "estimated_full_shards_per_hour": 70.00961776064356,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 7456296,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0635986559996127,
+      "bytes": 6186061,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -14,3 +14,4 @@
 {"step": 316400, "loss": 0.581691725179553, "lr": 0.0003, "grad_norm": 0.16576541745061205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46312342500, "tokens_per_second": 148523.02108168503, "data_wait_s": 1.0182000096392585e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 316500, "loss": 0.5566594488918781, "lr": 0.0003, "grad_norm": 0.15412083499869786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46327034564, "tokens_per_second": 148839.8568091833, "data_wait_s": 5.790000159322517e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 316600, "loss": 0.6161280572414398, "lr": 0.0003, "grad_norm": 0.16630223680457928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46341711617, "tokens_per_second": 149630.62656756822, "data_wait_s": 1.0258999736834085e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 316700, "loss": 0.5909538753330708, "lr": 0.0003, "grad_norm": 0.16705435691830953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46356316831, "tokens_per_second": 128492.15625667019, "data_wait_s": 1.4368999927683035e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:624db27b5d47613f0b911114da6fb7747cdf67fe2e18c00a6b6807d31cbb8be4
-size 34053
+oid sha256:241e37152ec6be5e2a5f80a8abfbc403884b5e0d08175925df7cfccda6db66f0
+size 36171