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": 77187.65313174498,
-      "required_pages_per_second_upper_estimate": 0.005340020943684552
+      "required_token_positions_per_second": 76715.3989735943,
+      "required_pages_per_second_upper_estimate": 0.005307349305243072
     },
     {
       "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": 9648.456641468123,
-      "required_pages_per_second_upper_estimate": 0.004039607313096669
+      "required_token_positions_per_second": 9589.424871699288,
+      "required_pages_per_second_upper_estimate": 0.004014891943817989
     },
     {
       "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": 19296.913282936246,
-      "required_pages_per_second_upper_estimate": 0.004019038143010721
+      "required_token_positions_per_second": 19178.849743398576,
+      "required_pages_per_second_upper_estimate": 0.003994448621269939
     },
     {
       "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": 28945.36992440437,
-      "required_pages_per_second_upper_estimate": 0.001735230729947258
+      "required_token_positions_per_second": 28768.274615097864,
+      "required_pages_per_second_upper_estimate": 0.001724614136563213
     },
     {
       "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": 28945.36992440437,
-      "required_pages_per_second_upper_estimate": 0.002250375194395442
+      "required_token_positions_per_second": 28768.274615097864,
+      "required_pages_per_second_upper_estimate": 0.0022366068130568032
     },
     {
       "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": 28945.36992440437,
-      "required_pages_per_second_upper_estimate": 0.3712801262734492
+      "required_token_positions_per_second": 28768.274615097864,
+      "required_pages_per_second_upper_estimate": 0.36900853779579357
     }
   ],
   "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": 0.08750313167954005,
-    "producer_p95_seconds": 0.320557426999585,
-    "gpu_update_seconds": 0.9551786718292885,
-    "producer_batches_per_second": 11.42816240751552,
-    "gpu_batches_per_second": 1.0469245487704126,
-    "producer_headroom": 10.915936989860079,
+    "producer_batch_seconds": 0.481376551078057,
+    "producer_p95_seconds": 0.32571637999717495,
+    "gpu_update_seconds": 0.961058679045356,
+    "producer_batches_per_second": 2.077375804368681,
+    "gpu_batches_per_second": 1.0405191918076484,
+    "producer_headroom": 1.9964800464273482,
     "sustainable": true,
-    "buffer_seconds": 60.17625632524518,
+    "buffer_seconds": 60.54669677985743,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462874.1932209936,
-    "upload_bytes_per_second": 4185822.474152754,
+    "generation_bytes_per_second": 405613.3249350797,
+    "upload_bytes_per_second": 4157483.3729018304,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.32202670547825,
+    "estimated_full_shards_per_hour": 87.03517733611388,
     "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": 1.0979482909970102,
-      "bytes": 1151793,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1090724049972778,
       "bytes": 6289086,
@@ -562,6 +556,12 @@
       "bytes": 4298944,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.740884680999443,
+      "bytes": 961930,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 875000, "loss": 0.5282732248306274, "lr": 3e-05, "grad_norm": 0.30791818889206707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128075402216, "tokens_per_second": 150694.60485398571, "data_wait_s": 1.2407999747665599e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875100, "loss": 0.4834609925746918, "lr": 3e-05, "grad_norm": 0.3024347114563917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128090022645, "tokens_per_second": 139551.35050221166, "data_wait_s": 1.2865999451605603e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875200, "loss": 0.5135159492492676, "lr": 3e-05, "grad_norm": 0.3100302680594288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128104481050, "tokens_per_second": 149996.80240268543, "data_wait_s": 8.998002158477902e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 875300, "loss": 0.5060501098632812, "lr": 3e-05, "grad_norm": 0.3257328238374927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128119024039, "tokens_per_second": 151256.37447600192, "data_wait_s": 1.0635001672198996e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:964457544950eaf9149157e9f110521e704af4e1639a822a3a5455fddb814f57
-size 698232
+oid sha256:afe4a945d450e89b35a188b059f601599fe86ee39ada9047e1df15531dff4041
+size 700350