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": 75703.9831219617,
-      "required_pages_per_second_upper_estimate": 0.006561829387020579
+      "required_token_positions_per_second": 75780.19990541549,
+      "required_pages_per_second_upper_estimate": 0.006568435664640681
     },
     {
       "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": 9462.997890245213,
-      "required_pages_per_second_upper_estimate": 0.0037749313428455454
+      "required_token_positions_per_second": 9472.524988176936,
+      "required_pages_per_second_upper_estimate": 0.0037787318446533174
     },
     {
       "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": 18925.995780490426,
-      "required_pages_per_second_upper_estimate": 0.004378196037667028
+      "required_token_positions_per_second": 18945.049976353872,
+      "required_pages_per_second_upper_estimate": 0.004382603890537642
     },
     {
       "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": 28388.99367073564,
-      "required_pages_per_second_upper_estimate": 0.0017287407690091277
+      "required_token_positions_per_second": 28417.574964530806,
+      "required_pages_per_second_upper_estimate": 0.001730481219846794
     },
     {
       "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": 28388.99367073564,
-      "required_pages_per_second_upper_estimate": 0.002185463318537423
+      "required_token_positions_per_second": 28417.574964530806,
+      "required_pages_per_second_upper_estimate": 0.0021876635856518598
     },
     {
       "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": 28388.99367073564,
-      "required_pages_per_second_upper_estimate": 0.3524962895406539
+      "required_token_positions_per_second": 28417.574964530806,
+      "required_pages_per_second_upper_estimate": 0.3528511735541528
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.22740758702343555,
-    "producer_p95_seconds": 0.31379283199930796,
-    "gpu_update_seconds": 0.973898558035205,
-    "producer_batches_per_second": 4.397390663561919,
-    "gpu_batches_per_second": 1.0268009863547323,
-    "producer_headroom": 4.2826124263604255,
+    "producer_batch_seconds": 0.4492049770547055,
+    "producer_p95_seconds": 0.23380935699969996,
+    "gpu_update_seconds": 0.9729190486700098,
+    "producer_batches_per_second": 2.226155209937082,
+    "gpu_batches_per_second": 1.0278347426407266,
+    "producer_headroom": 2.1658688090437717,
     "sustainable": true,
-    "buffer_seconds": 60.38171059818271,
+    "buffer_seconds": 60.3209810175406,
     "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": 446325.6246867401,
-    "upload_bytes_per_second": 5859113.273311393,
+    "generation_bytes_per_second": 386285.4674081516,
+    "upload_bytes_per_second": 5799553.831827062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.77108912898686,
+    "estimated_full_shards_per_hour": 82.887869040331,
     "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": 1.1632403840012557,
-      "bytes": 11705717,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.197896118999779,
       "bytes": 12261236,
@@ -562,6 +556,12 @@
       "bytes": 11422716,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.048131932002434,
+      "bytes": 6792716,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -159,3 +159,4 @@
 {"step": 330900, "loss": 0.6047540437430143, "lr": 0.0003, "grad_norm": 0.142320208239958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48434964339, "tokens_per_second": 149440.5948079516, "data_wait_s": 1.091599915525876e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331000, "loss": 0.5964784566313028, "lr": 0.0003, "grad_norm": 0.1577556859694088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48449730099, "tokens_per_second": 150677.61027192196, "data_wait_s": 1.105499904952012e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331100, "loss": 0.6439957339316607, "lr": 0.0003, "grad_norm": 0.16588003596327064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48464504517, "tokens_per_second": 149608.35486879697, "data_wait_s": 1.7828999261837453e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 331200, "loss": 0.5629096124321222, "lr": 0.0003, "grad_norm": 0.15336929508148103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48479199854, "tokens_per_second": 148608.9672753796, "data_wait_s": 1.3127999409334734e-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:bb92eff3728cce5afce0da2e2dfb24531225bf08241f1a18d21e2a0f7795f368
-size 343512
+oid sha256:78bd9f5b05df149f57367e49a8a7f10d23de8506d690fb632b91891a632d9f62
+size 345630