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": 77279.29290284136,
-      "required_pages_per_second_upper_estimate": 0.005346360795682564
+      "required_token_positions_per_second": 77179.03786147786,
+      "required_pages_per_second_upper_estimate": 0.005339424919297292
     },
     {
       "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": 9659.91161285517,
-      "required_pages_per_second_upper_estimate": 0.0040444032704094225
+      "required_token_positions_per_second": 9647.379732684733,
+      "required_pages_per_second_upper_estimate": 0.004039156433877476
     },
     {
       "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": 19319.82322571034,
-      "required_pages_per_second_upper_estimate": 0.004023809679914746
+      "required_token_positions_per_second": 19294.759465369465,
+      "required_pages_per_second_upper_estimate": 0.00401858955961155
     },
     {
       "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": 28979.734838565513,
-      "required_pages_per_second_upper_estimate": 0.0017372908540790328
+      "required_token_positions_per_second": 28942.139198054196,
+      "required_pages_per_second_upper_estimate": 0.0017350370528355986
     },
     {
       "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": 28979.734838565513,
-      "required_pages_per_second_upper_estimate": 0.0022530469153161877
+      "required_token_positions_per_second": 28942.139198054196,
+      "required_pages_per_second_upper_estimate": 0.002250124019632179
     },
     {
       "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": 28979.734838565513,
-      "required_pages_per_second_upper_estimate": 0.37172092249413824
+      "required_token_positions_per_second": 28942.139198054196,
+      "required_pages_per_second_upper_estimate": 0.3712386859847128
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.05984396739037834,
-    "producer_p95_seconds": 0.20638466799937305,
-    "gpu_update_seconds": 0.9540459964183912,
-    "producer_batches_per_second": 16.71012206588394,
-    "gpu_batches_per_second": 1.0481674927143199,
-    "producer_headroom": 15.942225056619188,
+    "producer_batch_seconds": 0.07854484391384631,
+    "producer_p95_seconds": 0.21246695999798249,
+    "gpu_update_seconds": 0.9552852956307666,
+    "producer_batches_per_second": 12.73158045990737,
+    "gpu_batches_per_second": 1.0468076966888815,
+    "producer_headroom": 12.162291603489503,
     "sustainable": true,
-    "buffer_seconds": 60.10489777435864,
+    "buffer_seconds": 60.182973624738295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360169.6136000602,
-    "upload_bytes_per_second": 5770036.829195146,
+    "generation_bytes_per_second": 360007.1691859602,
+    "upload_bytes_per_second": 5694545.773792408,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.28401475907664,
+    "estimated_full_shards_per_hour": 77.2491579693232,
     "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": 1.1187948159986263,
-      "bytes": 10435269,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 0.8605066399995849,
       "bytes": 757336,
@@ -562,6 +556,12 @@
       "bytes": 18642145,
       "pages": 45,
       "shards": 5
+    },
+    {
+      "seconds": 1.3427877739995893,
+      "bytes": 6044460,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -155,3 +155,4 @@
 {"step": 540500, "loss": 0.5647328495979309, "lr": 0.0003, "grad_norm": 0.15255013953858765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79110945736, "tokens_per_second": 149340.02235280143, "data_wait_s": 1.3532000593841076e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 540600, "loss": 0.5162097811698914, "lr": 0.0003, "grad_norm": 0.15450082272411939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79125690693, "tokens_per_second": 154192.46738125163, "data_wait_s": 1.1871001333929598e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 540700, "loss": 0.5719364881515503, "lr": 0.0003, "grad_norm": 0.15767005672336482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79140476289, "tokens_per_second": 154338.05389398625, "data_wait_s": 9.760999091668054e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 540800, "loss": 0.5983349680900574, "lr": 0.0003, "grad_norm": 0.15553669881718665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79155188870, "tokens_per_second": 153084.56724384284, "data_wait_s": 1.0845000360859558e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c63f4c06e4912c41f42480484c9a546a0723b58edcba35395493d6a61d473684
-size 335040
+oid sha256:c9b91220e4be5084ece53ce6e929fb85670bb90b249e93d99f05483b85247fd4
+size 337158