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": 77190.8286891098,
-      "required_pages_per_second_upper_estimate": 0.0053402406361113585
+      "required_token_positions_per_second": 77235.66766061819,
+      "required_pages_per_second_upper_estimate": 0.005343342700201057
     },
     {
       "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": 9648.853586138724,
-      "required_pages_per_second_upper_estimate": 0.004039773505541103
+      "required_token_positions_per_second": 9654.458457577273,
+      "required_pages_per_second_upper_estimate": 0.004042120148169398
     },
     {
       "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": 19297.70717227745,
-      "required_pages_per_second_upper_estimate": 0.004019203489224224
+      "required_token_positions_per_second": 19308.916915154547,
+      "required_pages_per_second_upper_estimate": 0.004021538183044724
     },
     {
       "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": 28946.56075841617,
-      "required_pages_per_second_upper_estimate": 0.001735302118627967
+      "required_token_positions_per_second": 28963.37537273182,
+      "required_pages_per_second_upper_estimate": 0.001736310129081761
     },
     {
       "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": 28946.56075841617,
-      "required_pages_per_second_upper_estimate": 0.002250467776501934
+      "required_token_positions_per_second": 28963.37537273182,
+      "required_pages_per_second_upper_estimate": 0.0022517750388053037
     },
     {
       "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": 28946.56075841617,
-      "required_pages_per_second_upper_estimate": 0.3712954010135346
+      "required_token_positions_per_second": 28963.37537273182,
+      "required_pages_per_second_upper_estimate": 0.3715110808318495
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08702350431249783,
-    "producer_p95_seconds": 0.25139905999822076,
-    "gpu_update_seconds": 0.9551393766861019,
-    "producer_batches_per_second": 11.491148373076784,
-    "gpu_batches_per_second": 1.046967620023733,
-    "producer_headroom": 10.975648294468074,
+    "producer_batch_seconds": 0.12906891888283667,
+    "producer_p95_seconds": 0.2666281350029749,
+    "gpu_update_seconds": 0.9545848729367984,
+    "producer_batches_per_second": 7.747798685040183,
+    "gpu_batches_per_second": 1.047575787497534,
+    "producer_headroom": 7.395931423298976,
     "sustainable": true,
-    "buffer_seconds": 60.17378073122442,
+    "buffer_seconds": 60.1388469950183,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313933.1266680158,
-    "upload_bytes_per_second": 3529547.5880852714,
+    "generation_bytes_per_second": 313734.99361403054,
+    "upload_bytes_per_second": 3546014.3278033827,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.36274099378925,
+    "estimated_full_shards_per_hour": 67.32022625270545,
     "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": 1.3536671069996373,
-      "bytes": 6151717,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1601749330002349,
       "bytes": 883373,
@@ -562,6 +556,12 @@
       "bytes": 9837963,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2818151600004057,
+      "bytes": 7876217,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 863700, "loss": 0.5926230549812317, "lr": 3e-05, "grad_norm": 0.3352663516243666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126421455380, "tokens_per_second": 152880.4483452441, "data_wait_s": 1.38890027301386e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 863800, "loss": 0.5399754643440247, "lr": 3e-05, "grad_norm": 0.3145661950903577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126436236550, "tokens_per_second": 153481.73047923108, "data_wait_s": 1.121400055126287e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 863900, "loss": 0.5542579889297485, "lr": 3e-05, "grad_norm": 0.3171998236379697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126450890686, "tokens_per_second": 151451.18440000675, "data_wait_s": 9.624000085750595e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 864000, "loss": 0.5007447004318237, "lr": 3e-05, "grad_norm": 0.3006990500875119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126465423623, "tokens_per_second": 151092.7277189139, "data_wait_s": 9.35500065679662e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:483819b8b7acc125ca7f3a1510695b5e333a403eddc15ec4842f0936ba999519
-size 456549
+oid sha256:f28e74b3cc1c70392ead9c9aca0e20446966b73eff76b41f8ee87935b8fc9280
+size 458667