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": 76211.37226200104,
-      "required_pages_per_second_upper_estimate": 0.005272479567831953
+      "required_token_positions_per_second": 77155.98983631944,
+      "required_pages_per_second_upper_estimate": 0.005337830403438046
     },
     {
       "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": 9526.42153275013,
-      "required_pages_per_second_upper_estimate": 0.003988513761459302
+      "required_token_positions_per_second": 9644.49872953993,
+      "required_pages_per_second_upper_estimate": 0.004037950218022934
     },
     {
       "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": 19052.84306550026,
-      "required_pages_per_second_upper_estimate": 0.00396820475328328
+      "required_token_positions_per_second": 19288.99745907986,
+      "required_pages_per_second_upper_estimate": 0.00401738948565575
     },
     {
       "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": 28579.26459825039,
-      "required_pages_per_second_upper_estimate": 0.001713283274654792
+      "required_token_positions_per_second": 28933.49618861979,
+      "required_pages_per_second_upper_estimate": 0.001734518917617117
     },
     {
       "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": 28579.26459825039,
-      "required_pages_per_second_upper_estimate": 0.0022219121156141146
+      "required_token_positions_per_second": 28933.49618861979,
+      "required_pages_per_second_upper_estimate": 0.002249452063665235
     },
     {
       "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": 28579.26459825039,
-      "required_pages_per_second_upper_estimate": 0.3665841202428187
+      "required_token_positions_per_second": 28933.49618861979,
+      "required_pages_per_second_upper_estimate": 0.37112782273982875
     }
   ],
   "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": 0.09247009889853075,
-    "producer_p95_seconds": 0.2715463140011707,
-    "gpu_update_seconds": 0.9674146759428022,
-    "producer_batches_per_second": 10.814306591121088,
-    "gpu_batches_per_second": 1.0336828920084777,
-    "producer_headroom": 10.461918906395518,
+    "producer_batch_seconds": 0.07190304428917216,
+    "producer_p95_seconds": 0.2541535679993103,
+    "gpu_update_seconds": 0.9555706583041491,
+    "producer_batches_per_second": 13.907616984592535,
+    "gpu_batches_per_second": 1.0464950878407042,
+    "producer_headroom": 13.289710717409054,
     "sustainable": true,
-    "buffer_seconds": 60.94712458439653,
+    "buffer_seconds": 60.200951473161396,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354105.9930772392,
-    "upload_bytes_per_second": 3230278.0044667525,
+    "generation_bytes_per_second": 352960.23280653014,
+    "upload_bytes_per_second": 3263103.8630459253,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.98290294874079,
+    "estimated_full_shards_per_hour": 75.73704946658066,
     "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": 1.6485553290003736,
-      "bytes": 917525,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 2.0703852710030333,
       "bytes": 13336105,
@@ -562,6 +556,12 @@
       "bytes": 2727332,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.913498541998706,
+      "bytes": 6177731,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -289,3 +289,4 @@
 {"step": 828900, "loss": 0.5660039782524109, "lr": 3e-05, "grad_norm": 0.29373261987504706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121327800196, "tokens_per_second": 152236.0218599373, "data_wait_s": 1.551900277263485e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829000, "loss": 0.5336809158325195, "lr": 3e-05, "grad_norm": 0.3212830800290023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121342467616, "tokens_per_second": 151047.79262992184, "data_wait_s": 1.3477001630235463e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829100, "loss": 0.5624409914016724, "lr": 3e-05, "grad_norm": 0.2926521655202817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121357081539, "tokens_per_second": 151903.7968278472, "data_wait_s": 8.317998435813934e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 829200, "loss": 0.5322349667549133, "lr": 3e-05, "grad_norm": 0.28532622734203844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121371716667, "tokens_per_second": 140601.0131580074, "data_wait_s": 7.474998710677028e-06, "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 sha256:8983ef5bfbeb630337c60754a0721be08ad71cef0d7d78a243f0121ad20a8e58
-size 620418
+oid sha256:c9dad97797a62b112dbcd8335a4a7feaf97163b353d2cec03ebd26f2adb60201
+size 622536