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": 77485.99721609925,
-      "required_pages_per_second_upper_estimate": 0.00536066107969383
+      "required_token_positions_per_second": 77553.66684781916,
+      "required_pages_per_second_upper_estimate": 0.0053653426218313914
     },
     {
       "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": 9685.749652012406,
-      "required_pages_per_second_upper_estimate": 0.004055221117844944
+      "required_token_positions_per_second": 9694.208355977395,
+      "required_pages_per_second_upper_estimate": 0.00405876260055726
     },
     {
       "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": 19371.499304024812,
-      "required_pages_per_second_upper_estimate": 0.004034572444237821
+      "required_token_positions_per_second": 19388.41671195479,
+      "required_pages_per_second_upper_estimate": 0.0040380958941675866
     },
     {
       "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": 29057.248956037216,
-      "required_pages_per_second_upper_estimate": 0.001741937707064257
+      "required_token_positions_per_second": 29082.625067932186,
+      "required_pages_per_second_upper_estimate": 0.0017434589662252812
     },
     {
       "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": 29057.248956037216,
-      "required_pages_per_second_upper_estimate": 0.0022590732970010426
+      "required_token_positions_per_second": 29082.625067932186,
+      "required_pages_per_second_upper_estimate": 0.0022610461780831564
     },
     {
       "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": 29057.248956037216,
-      "required_pages_per_second_upper_estimate": 0.37271519036489037
+      "required_token_positions_per_second": 29082.625067932186,
+      "required_pages_per_second_upper_estimate": 0.3730406878815329
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3697431128280755,
-    "producer_p95_seconds": 0.19571688999894832,
-    "gpu_update_seconds": 0.951500950479883,
-    "producer_batches_per_second": 2.704580464937514,
-    "gpu_batches_per_second": 1.0509710993937071,
-    "producer_headroom": 2.5734108830373685,
+    "producer_batch_seconds": 0.07040553828908003,
+    "producer_p95_seconds": 0.17762359399966954,
+    "gpu_update_seconds": 0.9506707161206687,
+    "producer_batches_per_second": 14.203428086780228,
+    "gpu_batches_per_second": 1.0518889275149084,
+    "producer_headroom": 13.502783150627778,
     "sustainable": true,
-    "buffer_seconds": 60.896060830712514,
+    "buffer_seconds": 60.842925831722795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324230.2238045508,
-    "upload_bytes_per_second": 3381864.4793595634,
+    "generation_bytes_per_second": 355833.8131230859,
+    "upload_bytes_per_second": 3409288.2780103944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.57225833513635,
+    "estimated_full_shards_per_hour": 76.35365290898736,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 6734300,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9863095519995113,
+      "bytes": 6175312,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -96,3 +96,4 @@
 {"step": 654600, "loss": 0.5277031064033508, "lr": 0.0001, "grad_norm": 0.18878068228114828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95816147784, "tokens_per_second": 150074.68669996143, "data_wait_s": 9.293999028159305e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 654700, "loss": 0.5652675628662109, "lr": 0.0001, "grad_norm": 0.17747562316914484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95830843449, "tokens_per_second": 153203.61626854952, "data_wait_s": 1.1227000868530013e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 654800, "loss": 0.5525216460227966, "lr": 0.0001, "grad_norm": 0.1807556533966403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95845512701, "tokens_per_second": 152991.05662446775, "data_wait_s": 8.007998985704035e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 654900, "loss": 0.5827817320823669, "lr": 0.0001, "grad_norm": 0.1737878096522394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95860166615, "tokens_per_second": 153401.84862541239, "data_wait_s": 8.524999429937452e-06, "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 sha256:463c4ab3edaa68d19986f6908e4932e4b7032d65dd99d087a4e587167f80b733
-size 208512
+oid sha256:ce2556602de3f449a29ef78a512e543613be6c95dfa71441e15aedb3902413ac
+size 210630