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": 75021.71285716708,
-      "required_pages_per_second_upper_estimate": 0.0065026919296663575
+      "required_token_positions_per_second": 73843.86885215543,
+      "required_pages_per_second_upper_estimate": 0.006400599396530286
     },
     {
       "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": 9377.714107145885,
-      "required_pages_per_second_upper_estimate": 0.003740910366661036
+      "required_token_positions_per_second": 9230.483606519429,
+      "required_pages_per_second_upper_estimate": 0.0036821779186689916
     },
     {
       "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": 18755.42821429177,
-      "required_pages_per_second_upper_estimate": 0.0043387382331136065
+      "required_token_positions_per_second": 18460.967213038857,
+      "required_pages_per_second_upper_estimate": 0.004270619862810369
     },
     {
       "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": 28133.142321437652,
-      "required_pages_per_second_upper_estimate": 0.001713160764185168
+      "required_token_positions_per_second": 27691.450819558286,
+      "required_pages_per_second_upper_estimate": 0.0016862640690969287
     },
     {
       "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": 28133.142321437652,
-      "required_pages_per_second_upper_estimate": 0.0021657671734265993
+      "required_token_positions_per_second": 27691.450819558286,
+      "required_pages_per_second_upper_estimate": 0.0021317645389315938
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28133.142321437652,
-      "required_pages_per_second_upper_estimate": 0.3493194720617561
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27691.450819558286,
+      "required_pages_per_second_upper_estimate": 0.34383514185477837
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.19619156360161583,
-    "producer_p95_seconds": 0.19941095000103815,
-    "gpu_update_seconds": 0.9827554876062327,
-    "producer_batches_per_second": 5.097059127530008,
-    "gpu_batches_per_second": 1.0175471036399615,
-    "producer_headroom": 5.009162828233552,
+    "producer_batch_seconds": 0.10607608259384449,
+    "producer_p95_seconds": 0.2160918280005717,
+    "gpu_update_seconds": 0.9984308940748025,
+    "producer_batches_per_second": 9.42719579708564,
+    "gpu_batches_per_second": 1.0015715718879588,
+    "producer_headroom": 9.412403528302436,
     "sustainable": true,
-    "buffer_seconds": 60.93084023158642,
+    "buffer_seconds": 60.90428453856295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333271.7368757514,
+    "generation_bytes_per_second": 391320.6513255723,
     "upload_bytes_per_second": 4967026.690318246,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.51235656456382,
+    "estimated_full_shards_per_hour": 83.96830229592682,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -121,3 +121,4 @@
 {"step": 249200, "loss": 0.6431687735021114, "lr": 0.0003, "grad_norm": 0.13891328064437083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36476268973, "tokens_per_second": 121536.00045341575, "data_wait_s": 1.3046999811194837e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249300, "loss": 0.5752866845577955, "lr": 0.0003, "grad_norm": 0.16293882183368283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36491026020, "tokens_per_second": 150820.65796480002, "data_wait_s": 1.0529000064707361e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249400, "loss": 0.6120243947952986, "lr": 0.0003, "grad_norm": 0.14405296487039118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36505670984, "tokens_per_second": 149597.77159190035, "data_wait_s": 1.2959999367012642e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 249500, "loss": 0.5780062265694141, "lr": 0.0003, "grad_norm": 0.1606750468806231, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36520295531, "tokens_per_second": 147778.30449327236, "data_wait_s": 8.953998985816725e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:faea2d217aa3063efb67bf5cb5e3341c19ff3ad4b1235a2854a1458741031c36
-size 262245
+oid sha256:2798aea89408e66be83ed811f372890da0283547a0f0defc1032ce1ec0ec8db5
+size 264363