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": 77265.08339386509,
-      "required_pages_per_second_upper_estimate": 0.0053453777488809825
+      "required_token_positions_per_second": 77655.98579727215,
+      "required_pages_per_second_upper_estimate": 0.0053724212841672215
     },
     {
       "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": 9658.135424233136,
-      "required_pages_per_second_upper_estimate": 0.004043659617324413
+      "required_token_positions_per_second": 9706.998224659019,
+      "required_pages_per_second_upper_estimate": 0.004064117451491427
     },
     {
       "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": 19316.270848466273,
-      "required_pages_per_second_upper_estimate": 0.004023069813417294
+      "required_token_positions_per_second": 19413.996449318038,
+      "required_pages_per_second_upper_estimate": 0.004043423478877313
     },
     {
       "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": 28974.40627269941,
-      "required_pages_per_second_upper_estimate": 0.0017369714146916089
+      "required_token_positions_per_second": 29120.994673977057,
+      "required_pages_per_second_upper_estimate": 0.001745759165520675
     },
     {
       "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": 28974.40627269941,
-      "required_pages_per_second_upper_estimate": 0.0022526326427579823
+      "required_token_positions_per_second": 29120.994673977057,
+      "required_pages_per_second_upper_estimate": 0.002264029246183084
     },
     {
       "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": 28974.40627269941,
-      "required_pages_per_second_upper_estimate": 0.3716525733725761
+      "required_token_positions_per_second": 29120.994673977057,
+      "required_pages_per_second_upper_estimate": 0.3735328519898033
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1347239279217547,
-    "producer_p95_seconds": 0.33090281799741206,
-    "gpu_update_seconds": 0.9542214511588045,
-    "producer_batches_per_second": 7.4225864360248055,
-    "gpu_batches_per_second": 1.0479747639141859,
-    "producer_headroom": 7.082791200335249,
+    "producer_batch_seconds": 0.12706199160922438,
+    "producer_p95_seconds": 0.30469649999940884,
+    "gpu_update_seconds": 0.9494181194540945,
+    "producer_batches_per_second": 7.870174135751564,
+    "gpu_batches_per_second": 1.0532767170853972,
+    "producer_headroom": 7.472085927741504,
     "sustainable": true,
-    "buffer_seconds": 60.11595142300469,
+    "buffer_seconds": 60.76275964506205,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360832.0259919361,
-    "upload_bytes_per_second": 5644154.588585353,
+    "generation_bytes_per_second": 346048.0102167679,
+    "upload_bytes_per_second": 5588276.00376454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.42615303820193,
+    "estimated_full_shards_per_hour": 74.25384740712431,
     "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": 1.4511087869977928,
-      "bytes": 13183448,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.3023088439986168,
       "bytes": 6313993,
@@ -562,6 +556,12 @@
       "bytes": 6376962,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2160047779980232,
+      "bytes": 7548487,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 675900, "loss": 0.5435974597930908, "lr": 0.0001, "grad_norm": 0.19091877172449642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98934190928, "tokens_per_second": 151262.43462264814, "data_wait_s": 1.371300095343031e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 676000, "loss": 0.5399375557899475, "lr": 0.0001, "grad_norm": 0.2007481949378553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98948936877, "tokens_per_second": 152763.95416736437, "data_wait_s": 1.2451997463358566e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 676100, "loss": 0.5532440543174744, "lr": 0.0001, "grad_norm": 0.2063494843930429, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98963527617, "tokens_per_second": 152708.34097409106, "data_wait_s": 8.895000064512715e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 676200, "loss": 0.5398682355880737, "lr": 0.0001, "grad_norm": 0.19113540890617703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98978267812, "tokens_per_second": 144062.50475073588, "data_wait_s": 1.7950998881133273e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:711ba3447ccb87ba2bfec2416202d1ed1a47df3bd7937a8b53b276c60a7519d9
-size 663561
+oid sha256:b5e67a95fca12dfdf5eab90950a76e234c42894f01ab199624343a6eca875e6a
+size 665679