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": 76732.74449624326,
-      "required_pages_per_second_upper_estimate": 0.005308549308747085
+      "required_token_positions_per_second": 77167.60010810602,
+      "required_pages_per_second_upper_estimate": 0.005338633628979784
     },
     {
       "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": 9591.593062030408,
-      "required_pages_per_second_upper_estimate": 0.004015799719832665
+      "required_token_positions_per_second": 9645.950013513253,
+      "required_pages_per_second_upper_estimate": 0.0040385578403163095
     },
     {
       "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": 19183.186124060816,
-      "required_pages_per_second_upper_estimate": 0.003995351775003836
+      "required_token_positions_per_second": 19291.900027026506,
+      "required_pages_per_second_upper_estimate": 0.004017994014013171
     },
     {
       "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": 28774.779186091222,
-      "required_pages_per_second_upper_estimate": 0.00172500407566236
+      "required_token_positions_per_second": 28937.850040539757,
+      "required_pages_per_second_upper_estimate": 0.0017347799243917722
     },
     {
       "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": 28774.779186091222,
-      "required_pages_per_second_upper_estimate": 0.0022371125148409527
+      "required_token_positions_per_second": 28937.850040539757,
+      "required_pages_per_second_upper_estimate": 0.0022497905565014415
     },
     {
       "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": 28774.779186091222,
-      "required_pages_per_second_upper_estimate": 0.3690919714484322
+      "required_token_positions_per_second": 28937.850040539757,
+      "required_pages_per_second_upper_estimate": 0.3711836692774561
     }
   ],
   "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": 0.060264174703092976,
-    "producer_p95_seconds": 0.20990857899960247,
-    "gpu_update_seconds": 0.9608414306568903,
-    "producer_batches_per_second": 16.59360648223855,
-    "gpu_batches_per_second": 1.0407544555154522,
-    "producer_headroom": 15.943824592151536,
+    "producer_batch_seconds": 0.122117198335701,
+    "producer_p95_seconds": 0.2156032620005135,
+    "gpu_update_seconds": 0.9554268876667488,
+    "producer_batches_per_second": 8.188854752882499,
+    "gpu_batches_per_second": 1.0466525622301706,
+    "producer_headroom": 7.823852010101589,
     "sustainable": true,
-    "buffer_seconds": 60.53301013138409,
+    "buffer_seconds": 60.191893923005175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363530.1802586509,
-    "upload_bytes_per_second": 3650767.8953550556,
+    "generation_bytes_per_second": 366988.3056559366,
+    "upload_bytes_per_second": 3571749.3829411617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.0051141340222,
+    "estimated_full_shards_per_hour": 78.74714734324048,
     "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": 1.3486207620007917,
-      "bytes": 11139901,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 0.9246388850006042,
       "bytes": 1100937,
@@ -562,6 +556,12 @@
       "bytes": 10890364,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.6343959760015423,
+      "bytes": 2597952,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -179,3 +179,4 @@
 {"step": 942900, "loss": 0.5169530510902405, "lr": 3e-05, "grad_norm": 0.3398005991344303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138014046013, "tokens_per_second": 143466.18323986366, "data_wait_s": 1.3903001672588289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 943000, "loss": 0.5173076391220093, "lr": 3e-05, "grad_norm": 0.33327081484553933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138028788788, "tokens_per_second": 153344.13158151816, "data_wait_s": 1.4615998225053772e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 943100, "loss": 0.5516357421875, "lr": 3e-05, "grad_norm": 0.3265654176664102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138043429903, "tokens_per_second": 152675.90930268308, "data_wait_s": 1.3322001905180514e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 943200, "loss": 0.5323226451873779, "lr": 3e-05, "grad_norm": 0.34239969708404827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138057994585, "tokens_per_second": 151893.92452867344, "data_wait_s": 1.3779997971141711e-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:adc73553d91a0d908d7677c1db45785a691dd41eaa3c7e9f1aca4369e005c69a
-size 385872
+oid sha256:525b44c14c8995fafed397f547c121472a41133a0dad531ee23648ccad27017f
+size 387990