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": 77310.24624976238,
-      "required_pages_per_second_upper_estimate": 0.00534850222004422
+      "required_token_positions_per_second": 77127.66550282805,
+      "required_pages_per_second_upper_estimate": 0.005335870860325603
     },
     {
       "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": 9663.780781220297,
-      "required_pages_per_second_upper_estimate": 0.004046023210406478
+      "required_token_positions_per_second": 9640.958187853506,
+      "required_pages_per_second_upper_estimate": 0.0040364678671537466
     },
     {
       "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": 19327.561562440595,
-      "required_pages_per_second_upper_estimate": 0.004025421371381994
+      "required_token_positions_per_second": 19281.916375707013,
+      "required_pages_per_second_upper_estimate": 0.0040159146827299114
     },
     {
       "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": 28991.34234366089,
-      "required_pages_per_second_upper_estimate": 0.0017379867062858666
+      "required_token_positions_per_second": 28922.87456356052,
+      "required_pages_per_second_upper_estimate": 0.0017338821674130961
     },
     {
       "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": 28991.34234366089,
-      "required_pages_per_second_upper_estimate": 0.002253949347781334
+      "required_token_positions_per_second": 28922.87456356052,
+      "required_pages_per_second_upper_estimate": 0.0022486262790364676
     },
     {
       "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": 28991.34234366089,
-      "required_pages_per_second_upper_estimate": 0.3718698111063338
+      "required_token_positions_per_second": 28922.87456356052,
+      "required_pages_per_second_upper_estimate": 0.3709915799381809
     }
   ],
   "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": 0.44901407733615883,
-    "producer_p95_seconds": 0.29027844599841046,
-    "gpu_update_seconds": 0.9536640170800984,
-    "producer_batches_per_second": 2.2271016666841383,
-    "gpu_batches_per_second": 1.0485873243511608,
-    "producer_headroom": 2.1239067218957777,
+    "producer_batch_seconds": 0.09006942649250504,
+    "producer_p95_seconds": 0.27771760600080597,
+    "gpu_update_seconds": 0.9559215816962151,
+    "producer_batches_per_second": 11.102546545949341,
+    "gpu_batches_per_second": 1.0461109144806322,
+    "producer_headroom": 10.613163855059744,
     "sustainable": true,
-    "buffer_seconds": 60.0808330760462,
+    "buffer_seconds": 60.22305964686155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329603.9945324498,
-    "upload_bytes_per_second": 3926243.389297781,
+    "generation_bytes_per_second": 328075.407828953,
+    "upload_bytes_per_second": 3915800.367332622,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.72534443836328,
+    "estimated_full_shards_per_hour": 70.39734531547015,
     "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": 0.908222386999114,
-      "bytes": 5648395,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3117666090001876,
       "bytes": 13264605,
@@ -562,6 +556,12 @@
       "bytes": 7316912,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2164157670013083,
+      "bytes": 5692225,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -277,3 +277,4 @@
 {"step": 552700, "loss": 0.5870258808135986, "lr": 0.0003, "grad_norm": 0.1753028595572003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80897426991, "tokens_per_second": 152592.72443008964, "data_wait_s": 1.2156000593677163e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552800, "loss": 0.5618245601654053, "lr": 0.0003, "grad_norm": 0.14908982222349665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80912163186, "tokens_per_second": 149433.32025715156, "data_wait_s": 1.841299672378227e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552900, "loss": 0.5635361671447754, "lr": 0.0003, "grad_norm": 0.15189729263302537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80926877470, "tokens_per_second": 152746.83972867098, "data_wait_s": 1.4863999240333214e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 553000, "loss": 0.5958641767501831, "lr": 0.0003, "grad_norm": 0.16391493848202127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80941456564, "tokens_per_second": 151671.7997913953, "data_wait_s": 1.3291999493958429e-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:cb497ed34c4eaf23afc6fce63fb1752bbe487e34d9ff67392d9798f5976ea6bb
-size 595002
+oid sha256:ab653af83ac60f52b02268fb4758a30df269b48ff5533efb6c6a8a4cbd1a2afc
+size 597120