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": 77193.63551302586,
-      "required_pages_per_second_upper_estimate": 0.0053404348187026
+      "required_token_positions_per_second": 77173.84471553985,
+      "required_pages_per_second_upper_estimate": 0.0053390656454115415
     },
     {
       "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": 9649.204439128232,
-      "required_pages_per_second_upper_estimate": 0.004039920400361166
+      "required_token_positions_per_second": 9646.73058944248,
+      "required_pages_per_second_upper_estimate": 0.004038884651157598
     },
     {
       "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": 19298.408878256465,
-      "required_pages_per_second_upper_estimate": 0.0040193496360744225
+      "required_token_positions_per_second": 19293.46117888496,
+      "required_pages_per_second_upper_estimate": 0.004018319160774945
     },
     {
       "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": 28947.613317384697,
-      "required_pages_per_second_upper_estimate": 0.0017353652179309658
+      "required_token_positions_per_second": 28940.19176832744,
+      "required_pages_per_second_upper_estimate": 0.0017349203073970326
     },
     {
       "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": 28947.613317384697,
-      "required_pages_per_second_upper_estimate": 0.0022505496083320194
+      "required_token_positions_per_second": 28940.19176832744,
+      "required_pages_per_second_upper_estimate": 0.002249972615536762
     },
     {
       "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": 28947.613317384697,
-      "required_pages_per_second_upper_estimate": 0.3713089021098331
+      "required_token_positions_per_second": 28940.19176832744,
+      "required_pages_per_second_upper_estimate": 0.3712137064471651
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06988396339065162,
-    "producer_p95_seconds": 0.18457786799990572,
-    "gpu_update_seconds": 0.9551046470347797,
-    "producer_batches_per_second": 14.309434546664107,
-    "gpu_batches_per_second": 1.0470056900095739,
-    "producer_headroom": 13.667007431958906,
+    "producer_batch_seconds": 0.14871029841408756,
+    "producer_p95_seconds": 0.19222498999988602,
+    "gpu_update_seconds": 0.9553495782380532,
+    "producer_batches_per_second": 6.724483849904429,
+    "gpu_batches_per_second": 1.046737260139158,
+    "producer_headroom": 6.4242328098747965,
     "sustainable": true,
-    "buffer_seconds": 60.17159276319112,
+    "buffer_seconds": 60.18702342899735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482175.2257147927,
-    "upload_bytes_per_second": 5876271.782359014,
+    "generation_bytes_per_second": 457628.04225284664,
+    "upload_bytes_per_second": 5858357.337815507,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.46357897360645,
+    "estimated_full_shards_per_hour": 98.19632483185815,
     "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": 1.079906239000593,
-      "bytes": 5671187,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1921040119996178,
       "bytes": 5850261,
@@ -562,6 +556,12 @@
       "bytes": 8273419,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3893709189997026,
+      "bytes": 6184789,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 535900, "loss": 0.5594803690910339, "lr": 0.0003, "grad_norm": 0.1546807488656789, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78438554401, "tokens_per_second": 155480.88742217954, "data_wait_s": 1.7360000128974207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 536000, "loss": 0.6030716300010681, "lr": 0.0003, "grad_norm": 0.1690846280061718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78453199370, "tokens_per_second": 153034.85031258326, "data_wait_s": 9.528999726171605e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 536100, "loss": 0.5438624024391174, "lr": 0.0003, "grad_norm": 0.155175211899847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78467636456, "tokens_per_second": 151072.1523831253, "data_wait_s": 1.2328999218880199e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 536200, "loss": 0.5933921933174133, "lr": 0.0003, "grad_norm": 0.1613617432365788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78482185020, "tokens_per_second": 152032.92833788562, "data_wait_s": 1.3897999451728538e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c834ab134ceba77873a64200c0aed864636e037d18fd740086ed569247a29aff
-size 236829
+oid sha256:51617d98b862e605f1d23cac363f7004d901874d31a1130a5919ecb3e991c186
+size 238947