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": 75703.8137889036,
-      "required_pages_per_second_upper_estimate": 0.006561814709660273
+      "required_token_positions_per_second": 75247.81803639756,
+      "required_pages_per_second_upper_estimate": 0.006522290153543718
     },
     {
       "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": 9462.97672361295,
-      "required_pages_per_second_upper_estimate": 0.0037749228991594665
+      "required_token_positions_per_second": 9405.977254549694,
+      "required_pages_per_second_upper_estimate": 0.0037521849587321262
     },
     {
       "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": 18925.9534472259,
-      "required_pages_per_second_upper_estimate": 0.00437818624461132
+      "required_token_positions_per_second": 18811.95450909939,
+      "required_pages_per_second_upper_estimate": 0.004351814596588539
     },
     {
       "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": 28388.93017083885,
-      "required_pages_per_second_upper_estimate": 0.0017287369021985715
+      "required_token_positions_per_second": 28217.931763649085,
+      "required_pages_per_second_upper_estimate": 0.0017183239963600213
     },
     {
       "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": 28388.93017083885,
-      "required_pages_per_second_upper_estimate": 0.002185458430139589
+      "required_token_positions_per_second": 28217.931763649085,
+      "required_pages_per_second_upper_estimate": 0.002172294499400234
     },
     {
       "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": 28388.93017083885,
-      "required_pages_per_second_upper_estimate": 0.3524955010844562
+      "required_token_positions_per_second": 28217.931763649085,
+      "required_pages_per_second_upper_estimate": 0.3503722731620135
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10452180185941984,
-    "producer_p95_seconds": 0.3406817700015381,
-    "gpu_update_seconds": 0.973900736435643,
-    "producer_batches_per_second": 9.567381945299642,
-    "gpu_batches_per_second": 1.0267986896281411,
-    "producer_headroom": 9.317680322288396,
+    "producer_batch_seconds": 0.3937103602970353,
+    "producer_p95_seconds": 0.3487115170028119,
+    "gpu_update_seconds": 0.9798024969220713,
+    "producer_batches_per_second": 2.5399382410093265,
+    "gpu_batches_per_second": 1.0206138514051317,
+    "producer_headroom": 2.4886378305687917,
     "sustainable": true,
-    "buffer_seconds": 60.38184565900987,
+    "buffer_seconds": 60.74775480916842,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299548.84641915036,
-    "upload_bytes_per_second": 3925264.591360211,
+    "generation_bytes_per_second": 281954.03522981086,
+    "upload_bytes_per_second": 3920641.0333583304,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.2762093012894,
+    "estimated_full_shards_per_hour": 60.50077240629906,
     "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": 1.1223962459989707,
-      "bytes": 6954584,
-      "pages": 39,
-      "shards": 2
-    },
     {
       "seconds": 1.6785005409983569,
       "bytes": 5701009,
@@ -562,6 +556,12 @@
       "bytes": 7935280,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.982757900001161,
+      "bytes": 5911177,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -255,3 +255,4 @@
 {"step": 140500, "loss": 0.5549229513853788, "lr": 0.0003, "grad_norm": 0.12578476302069969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20561851641, "tokens_per_second": 149664.62252536995, "data_wait_s": 1.4966997696319595e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140600, "loss": 0.5978204291313887, "lr": 0.0003, "grad_norm": 0.13984901672717093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20576622327, "tokens_per_second": 150139.21753891415, "data_wait_s": 1.62090000230819e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140700, "loss": 0.62378946878016, "lr": 0.0003, "grad_norm": 0.13608952735719315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20591427571, "tokens_per_second": 150778.95619423408, "data_wait_s": 1.3404001947492361e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 140800, "loss": 0.5893173590302467, "lr": 0.0003, "grad_norm": 0.13309273764161786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20606050889, "tokens_per_second": 149140.1489081394, "data_wait_s": 1.3107997801853344e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aadff28ea11e480eacbc188f806b6de3b20c69c9b386b5fc95f4786167de0380
-size 548406
+oid sha256:2f84f9d7404a9f4d37eac7e6bcce7dc2d663a2d372c834246020077cfcbe344c
+size 550524