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": 77381.49949206629,
-      "required_pages_per_second_upper_estimate": 0.005353431684677105
+      "required_token_positions_per_second": 77311.60272956933,
+      "required_pages_per_second_upper_estimate": 0.005348596064464732
     },
     {
       "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": 9672.687436508286,
-      "required_pages_per_second_upper_estimate": 0.004049752240983446
+      "required_token_positions_per_second": 9663.950341196167,
+      "required_pages_per_second_upper_estimate": 0.004046094201627559
     },
     {
       "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": 19345.37487301657,
-      "required_pages_per_second_upper_estimate": 0.00402913141420638
+      "required_token_positions_per_second": 19327.900682392334,
+      "required_pages_per_second_upper_estimate": 0.004025492001124747
     },
     {
       "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": 29018.06230952486,
-      "required_pages_per_second_upper_estimate": 0.001739588527440386
+      "required_token_positions_per_second": 28991.8510235885,
+      "required_pages_per_second_upper_estimate": 0.0017380172008708175
     },
     {
       "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": 29018.06230952486,
-      "required_pages_per_second_upper_estimate": 0.0022560267075985487
+      "required_token_positions_per_second": 28991.8510235885,
+      "required_pages_per_second_upper_estimate": 0.0022539888954082594
     },
     {
       "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": 29018.06230952486,
-      "required_pages_per_second_upper_estimate": 0.3722125461387727
+      "required_token_positions_per_second": 28991.8510235885,
+      "required_pages_per_second_upper_estimate": 0.37187633590626723
     }
   ],
   "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": 0.1370279940001069,
-    "producer_p95_seconds": 0.2955458850010473,
-    "gpu_update_seconds": 0.9527858788463918,
-    "producer_batches_per_second": 7.297778875747242,
-    "gpu_batches_per_second": 1.0495537583016803,
-    "producer_headroom": 6.953220659755469,
+    "producer_batch_seconds": 0.07815655659382514,
+    "producer_p95_seconds": 0.2871851029995014,
+    "gpu_update_seconds": 0.9536472844560663,
+    "producer_batches_per_second": 12.794831855207478,
+    "gpu_batches_per_second": 1.0486057227860424,
+    "producer_headroom": 12.201756653790584,
     "sustainable": true,
-    "buffer_seconds": 60.02551036732268,
+    "buffer_seconds": 60.079778920732174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306859.7806318285,
-    "upload_bytes_per_second": 5192301.786029192,
+    "generation_bytes_per_second": 307049.90689017467,
+    "upload_bytes_per_second": 5181004.5318630515,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.84496559349193,
+    "estimated_full_shards_per_hour": 65.88576226262026,
     "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": 1.0667447009982425,
-      "bytes": 2533809,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2533954940008698,
       "bytes": 6219875,
@@ -562,6 +556,12 @@
       "bytes": 11497273,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.0184783190015878,
+      "bytes": 1341723,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -268,3 +268,4 @@
 {"step": 951800, "loss": 0.5045588612556458, "lr": 3e-05, "grad_norm": 0.35447065909065006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139317685038, "tokens_per_second": 149707.28866007438, "data_wait_s": 1.1185002222191542e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951900, "loss": 0.569692850112915, "lr": 3e-05, "grad_norm": 0.3484491910040527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139332171594, "tokens_per_second": 150889.51401976566, "data_wait_s": 1.8131002434529364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952000, "loss": 0.5298699736595154, "lr": 3e-05, "grad_norm": 0.36111805114874257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139346666069, "tokens_per_second": 150348.00721994363, "data_wait_s": 1.2268999853404239e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 952100, "loss": 0.5685331225395203, "lr": 3e-05, "grad_norm": 0.5033133254657745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139361327151, "tokens_per_second": 152078.57846030113, "data_wait_s": 1.6996997146634385e-05, "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:eb19c67bee0d7ad12662044e2cc9f3255d81a256e9c7ca60b3b3541eba4c28fe
-size 575940
+oid sha256:ed26e4f8ff8ec564769a1cc0f6944fcac511d896ae18956a457120ee22a5c158
+size 578058