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": 76675.44066927915,
-      "required_pages_per_second_upper_estimate": 0.00530458489703451
+      "required_token_positions_per_second": 77231.39524893338,
+      "required_pages_per_second_upper_estimate": 0.005343047124847341
     },
     {
       "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": 9584.430083659894,
-      "required_pages_per_second_upper_estimate": 0.004012800730368929
+      "required_token_positions_per_second": 9653.924406116672,
+      "required_pages_per_second_upper_estimate": 0.004041896551975107
     },
     {
       "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": 19168.860167319788,
-      "required_pages_per_second_upper_estimate": 0.003992368056015565
+      "required_token_positions_per_second": 19307.848812233344,
+      "required_pages_per_second_upper_estimate": 0.004021315725374006
     },
     {
       "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": 28753.290250979684,
-      "required_pages_per_second_upper_estimate": 0.0017237158468141273
+      "required_token_positions_per_second": 28961.77321835002,
+      "required_pages_per_second_upper_estimate": 0.00173621408237292
     },
     {
       "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": 28753.290250979684,
-      "required_pages_per_second_upper_estimate": 0.0022354418446558673
+      "required_token_positions_per_second": 28961.77321835002,
+      "required_pages_per_second_upper_estimate": 0.002251650478349251
     },
     {
       "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": 28753.290250979684,
-      "required_pages_per_second_upper_estimate": 0.36881633446184225
+      "required_token_positions_per_second": 28961.77321835002,
+      "required_pages_per_second_upper_estimate": 0.3714905301156991
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.041392753664050375,
-    "producer_p95_seconds": 0.12836806399991474,
-    "gpu_update_seconds": 0.9615595209684907,
-    "producer_batches_per_second": 24.158817944709497,
-    "gpu_batches_per_second": 1.0399772226193462,
-    "producer_headroom": 23.230141410079842,
+    "producer_batch_seconds": 0.18028673940625062,
+    "producer_p95_seconds": 0.14007191299992883,
+    "gpu_update_seconds": 0.9546376802122869,
+    "producer_batches_per_second": 5.546719649450433,
+    "gpu_batches_per_second": 1.0475178392053681,
+    "producer_headroom": 5.2951075789392705,
     "sustainable": true,
-    "buffer_seconds": 60.578249821014914,
+    "buffer_seconds": 60.142173853374075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 69,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353908.15188023093,
-    "upload_bytes_per_second": 5564081.131730167,
+    "generation_bytes_per_second": 396238.04085439944,
+    "upload_bytes_per_second": 6192889.074439749,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.94045083337016,
+    "estimated_full_shards_per_hour": 85.02345961784351,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 6129536,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3353518999999778,
+      "bytes": 13219093,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -9,3 +9,4 @@
 {"step": 441100, "loss": 0.6048697233200073, "lr": 0.0003, "grad_norm": 0.1563407855641162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64563846639, "tokens_per_second": 145515.66358179986, "data_wait_s": 1.3674000001628883e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441200, "loss": 0.5839400291442871, "lr": 0.0003, "grad_norm": 0.1543983455049739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64578388517, "tokens_per_second": 151346.7096893474, "data_wait_s": 1.3643000102092628e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441300, "loss": 0.5678753852844238, "lr": 0.0003, "grad_norm": 0.16825666745203535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64593027227, "tokens_per_second": 152015.49152838832, "data_wait_s": 9.307000027547474e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 441400, "loss": 0.5791196227073669, "lr": 0.0003, "grad_norm": 0.14784625682046532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64607772031, "tokens_per_second": 152950.68091202807, "data_wait_s": 1.7840000055002747e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c1e8636708e996e6e3ad4332a1784b4beba4cc0952d753e10f1275a30865384
-size 23463
+oid sha256:c986000b5ddcb4bc2e0293018d2800345fb80c6fd87302cbaf903d3856dad214
+size 25581