Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75594.79450501325,
-      "required_pages_per_second_upper_estimate": 0.006552365194439506
+      "required_token_positions_per_second": 76041.00971871785,
+      "required_pages_per_second_upper_estimate": 0.006591041998241301
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9449.349313126657,
-      "required_pages_per_second_upper_estimate": 0.0037694867213685403
+      "required_token_positions_per_second": 9505.12621483973,
+      "required_pages_per_second_upper_estimate": 0.003791736961400882
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18898.698626253314,
-      "required_pages_per_second_upper_estimate": 0.004371881321447778
+      "required_token_positions_per_second": 19010.25242967946,
+      "required_pages_per_second_upper_estimate": 0.0043976873305905315
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28348.04793937997,
-      "required_pages_per_second_upper_estimate": 0.0017262473887952174
+      "required_token_positions_per_second": 28515.37864451919,
+      "required_pages_per_second_upper_estimate": 0.001736436950821835
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28348.04793937997,
-      "required_pages_per_second_upper_estimate": 0.0021823112027926244
+      "required_token_positions_per_second": 28515.37864451919,
+      "required_pages_per_second_upper_estimate": 0.00219519278367528
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28348.04793937997,
-      "required_pages_per_second_upper_estimate": 0.35198788059376396
+      "required_token_positions_per_second": 28515.37864451919,
+      "required_pages_per_second_upper_estimate": 0.35406556793174804
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07754010574998915,
-    "producer_p95_seconds": 0.19902874899889866,
-    "gpu_update_seconds": 0.9753052506163047,
-    "producer_batches_per_second": 12.89655192403629,
-    "gpu_batches_per_second": 1.0253200209555833,
-    "producer_headroom": 12.5780748063584,
+    "producer_batch_seconds": 0.08208364703125426,
+    "producer_p95_seconds": 0.2750826979990961,
+    "gpu_update_seconds": 0.969582075155579,
+    "producer_batches_per_second": 12.18269455814066,
+    "gpu_batches_per_second": 1.0313722021310472,
+    "producer_headroom": 11.8121222706686,
     "sustainable": true,
-    "buffer_seconds": 60.46892553821089,
+    "buffer_seconds": 60.1140886596459,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424543.194599225,
-    "upload_bytes_per_second": 3552286.506863147,
+    "generation_bytes_per_second": 406758.03619743575,
+    "upload_bytes_per_second": 3542780.4776496445,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.09708670122683,
+    "estimated_full_shards_per_hour": 87.2808057255011,
     "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": 1.26741677300015,
-      "bytes": 7410356,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.306584923000628,
       "bytes": 936913,
@@ -562,6 +556,12 @@
       "bytes": 1084848,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.2088771979997546,
+      "bytes": 6082441,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 248900, "loss": 0.5806682482361794, "lr": 0.0003, "grad_norm": 0.15996525742501994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36432411760, "tokens_per_second": 148240.40482222656, "data_wait_s": 1.6429999959655106e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249000, "loss": 0.5690943282097578, "lr": 0.0003, "grad_norm": 0.13730764362207964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36446962241, "tokens_per_second": 148038.11439091066, "data_wait_s": 1.4449999071075581e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249100, "loss": 0.5791001301258802, "lr": 0.0003, "grad_norm": 0.15355182679166174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36461575458, "tokens_per_second": 149025.73628882563, "data_wait_s": 1.1598000128287822e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 249200, "loss": 0.6431687735021114, "lr": 0.0003, "grad_norm": 0.13891328064437083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36476268973, "tokens_per_second": 121536.00045341575, "data_wait_s": 1.3046999811194837e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d82e11949c21442cce192d3c1bcc5229e2c77ff4d0ec28fd4e6ed19b9f654e90
-size 255891
+oid sha256:470d934fd698990f77bb5eae07642c3c8576d915adce6172fd3377666ceb2e44
+size 258009