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": 77013.07684325197,
-      "required_pages_per_second_upper_estimate": 0.005327943350973808
+      "required_token_positions_per_second": 77118.04447846866,
+      "required_pages_per_second_upper_estimate": 0.0053352052555365684
     },
     {
       "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": 9626.634605406496,
-      "required_pages_per_second_upper_estimate": 0.004030470882293598
+      "required_token_positions_per_second": 9639.755559808582,
+      "required_pages_per_second_upper_estimate": 0.004035964351905066
     },
     {
       "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": 19253.269210812992,
-      "required_pages_per_second_upper_estimate": 0.004009948233759029
+      "required_token_positions_per_second": 19279.511119617164,
+      "required_pages_per_second_upper_estimate": 0.004015413731317265
     },
     {
       "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": 28879.90381621949,
-      "required_pages_per_second_upper_estimate": 0.0017313061367225336
+      "required_token_positions_per_second": 28919.266679425746,
+      "required_pages_per_second_upper_estimate": 0.0017336658802681367
     },
     {
       "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": 28879.90381621949,
-      "required_pages_per_second_upper_estimate": 0.002245285492439045
+      "required_token_positions_per_second": 28919.266679425746,
+      "required_pages_per_second_upper_estimate": 0.002248345781914394
     },
     {
       "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": 28879.90381621949,
-      "required_pages_per_second_upper_estimate": 0.3704403973296839
+      "required_token_positions_per_second": 28919.266679425746,
+      "required_pages_per_second_upper_estimate": 0.37094530187434416
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11606190647654557,
+    "producer_batch_seconds": 0.3682769866638296,
     "producer_p95_seconds": 0.3368981410021661,
-    "gpu_update_seconds": 0.9573439086203733,
-    "producer_batches_per_second": 8.616091449454911,
-    "gpu_batches_per_second": 1.044556706315809,
-    "producer_headroom": 8.248562665251743,
+    "gpu_update_seconds": 0.9560408397101517,
+    "producer_batches_per_second": 2.7153475134541045,
+    "gpu_batches_per_second": 1.0459804209861743,
+    "producer_headroom": 2.5959831168675347,
     "sustainable": true,
-    "buffer_seconds": 60.31266624308352,
+    "buffer_seconds": 60.23057290173956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360863.91206177394,
+    "generation_bytes_per_second": 339294.10913994256,
     "upload_bytes_per_second": 4111518.735549046,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.43299504651941,
+    "estimated_full_shards_per_hour": 72.80461745880802,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 876300, "loss": 0.5127866268157959, "lr": 3e-05, "grad_norm": 0.3174092561154268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128265649775, "tokens_per_second": 152528.61273359173, "data_wait_s": 9.88500687526539e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 876400, "loss": 0.5109633207321167, "lr": 3e-05, "grad_norm": 0.3099059393556759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128280329031, "tokens_per_second": 152866.22827241596, "data_wait_s": 1.052000152412802e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 876500, "loss": 0.5438387393951416, "lr": 3e-05, "grad_norm": 0.34324478302689776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128294942161, "tokens_per_second": 151727.92724023445, "data_wait_s": 1.4151999494060874e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 876600, "loss": 0.5262576937675476, "lr": 3e-05, "grad_norm": 0.29965560832321864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128309628614, "tokens_per_second": 152601.63673376132, "data_wait_s": 1.5012003132142127e-05, "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 sha256:9a6df45515fa018171ed42d39b50e58755e7f9f8479b414d08a163cdb5830c8d
-size 725766
+oid sha256:98fdc047d1d90a3af43d5ff85aa4467cc49c3e738583d433815231752c3239aa
+size 727884