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": 76894.96215411894,
-      "required_pages_per_second_upper_estimate": 0.005319771902715748
+      "required_token_positions_per_second": 77298.32945005053,
+      "required_pages_per_second_upper_estimate": 0.005347677788189369
     },
     {
       "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": 9611.870269264868,
-      "required_pages_per_second_upper_estimate": 0.004024289363065496
+      "required_token_positions_per_second": 9662.291181256316,
+      "required_pages_per_second_upper_estimate": 0.004045399546007944
     },
     {
       "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": 19223.740538529735,
-      "required_pages_per_second_upper_estimate": 0.004003798190045882
+      "required_token_positions_per_second": 19324.582362512632,
+      "required_pages_per_second_upper_estimate": 0.0040248008826037854
     },
     {
       "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": 28835.610807794605,
-      "required_pages_per_second_upper_estimate": 0.0017286508384989697
+      "required_token_positions_per_second": 28986.873543768947,
+      "required_pages_per_second_upper_estimate": 0.0017377188085558068
     },
     {
       "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": 28835.610807794605,
-      "required_pages_per_second_upper_estimate": 0.0022418419058583645
+      "required_token_positions_per_second": 28986.873543768947,
+      "required_pages_per_second_upper_estimate": 0.002253601918245909
     },
     {
       "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": 28835.610807794605,
-      "required_pages_per_second_upper_estimate": 0.36987225417573666
+      "required_token_positions_per_second": 28986.873543768947,
+      "required_pages_per_second_upper_estimate": 0.37181249013954343
     }
   ],
   "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": 0.14993574045314517,
-    "producer_p95_seconds": 0.3690605240044533,
-    "gpu_update_seconds": 0.95881443900354,
-    "producer_batches_per_second": 6.669523870544391,
-    "gpu_batches_per_second": 1.0429546733143304,
-    "producer_headroom": 6.3948357883567395,
+    "producer_batch_seconds": 0.26229020336722897,
+    "producer_p95_seconds": 0.3767514140054118,
+    "gpu_update_seconds": 0.9538110399609911,
+    "producer_batches_per_second": 3.8125709125319998,
+    "gpu_batches_per_second": 1.0484256924106246,
+    "producer_headroom": 3.636472227007172,
     "sustainable": true,
-    "buffer_seconds": 60.40530965722302,
+    "buffer_seconds": 60.090095517542444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 480021.5600419657,
-    "upload_bytes_per_second": 4977250.725159401,
+    "generation_bytes_per_second": 470039.85887999873,
+    "upload_bytes_per_second": 5043200.50054794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.00145245498875,
+    "estimated_full_shards_per_hour": 100.85961174774143,
     "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": 1.0372320719980053,
-      "bytes": 2714870,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.8995327459997497,
       "bytes": 6052441,
@@ -562,6 +556,12 @@
       "bytes": 1313572,
       "pages": 37,
       "shards": 1
+    },
+    {
+      "seconds": 2.0110623340005986,
+      "bytes": 13186052,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 966600, "loss": 0.5612720251083374, "lr": 3e-05, "grad_norm": 0.34865132573195956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141483868462, "tokens_per_second": 151150.28644634155, "data_wait_s": 1.2174001312814653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 966700, "loss": 0.56767737865448, "lr": 3e-05, "grad_norm": 0.3578803120532856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141498646124, "tokens_per_second": 153726.08751925977, "data_wait_s": 1.355900167254731e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 966800, "loss": 0.5247310996055603, "lr": 3e-05, "grad_norm": 0.34686408593857804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141513344383, "tokens_per_second": 153019.54933956327, "data_wait_s": 1.104700641008094e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 966900, "loss": 0.5069453716278076, "lr": 3e-05, "grad_norm": 0.3382227076291205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141528113565, "tokens_per_second": 153223.04782701455, "data_wait_s": 9.103998309001327e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b25bfbe78779b4eaf7bd0617d5ecdcf75f68054e3ad2696fc22f127d6edecfb
-size 891753
+oid sha256:41e25003a568f04794da59524ba74ada66603b1a6d0f14d50c2847d684d90396
+size 893871