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": 75447.23037544634,
-      "required_pages_per_second_upper_estimate": 0.006539574709686513
+      "required_token_positions_per_second": 75974.51560188636,
+      "required_pages_per_second_upper_estimate": 0.006585278456722146
     },
     {
       "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": 9430.903796930792,
-      "required_pages_per_second_upper_estimate": 0.0037621285291729663
+      "required_token_positions_per_second": 9496.814450235795,
+      "required_pages_per_second_upper_estimate": 0.003788421274228417
     },
     {
       "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": 18861.807593861584,
-      "required_pages_per_second_upper_estimate": 0.004363347230363947
+      "required_token_positions_per_second": 18993.62890047159,
+      "required_pages_per_second_upper_estimate": 0.0043938417696724125
     },
     {
       "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": 28292.711390792378,
-      "required_pages_per_second_upper_estimate": 0.0017228776833146657
+      "required_token_positions_per_second": 28490.443350707385,
+      "required_pages_per_second_upper_estimate": 0.001734918522254072
     },
     {
       "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": 28292.711390792378,
-      "required_pages_per_second_upper_estimate": 0.002178051242100981
+      "required_token_positions_per_second": 28490.443350707385,
+      "required_pages_per_second_upper_estimate": 0.002193273195732336
     },
     {
       "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": 28292.711390792378,
-      "required_pages_per_second_upper_estimate": 0.35130078585982066
+      "required_token_positions_per_second": 28490.443350707385,
+      "required_pages_per_second_upper_estimate": 0.3537559550356654
     }
   ],
   "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": 0.1190258759218068,
-    "producer_p95_seconds": 0.3267332479990728,
-    "gpu_update_seconds": 0.9772128099747205,
-    "producer_batches_per_second": 8.401534475217328,
-    "gpu_batches_per_second": 1.0233185543544696,
-    "producer_headroom": 8.210087112626615,
+    "producer_batch_seconds": 0.07331303923422183,
+    "producer_p95_seconds": 0.309553130002314,
+    "gpu_update_seconds": 0.9704306689672322,
+    "producer_batches_per_second": 13.640138377092535,
+    "gpu_batches_per_second": 1.0304703179509327,
+    "producer_headroom": 13.236808610087527,
     "sustainable": true,
-    "buffer_seconds": 60.58719421843267,
+    "buffer_seconds": 60.166701475968395,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364600.3004160497,
-    "upload_bytes_per_second": 2278378.7688018265,
+    "generation_bytes_per_second": 364509.98805771617,
+    "upload_bytes_per_second": 2252856.255260649,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.23473700867766,
+    "estimated_full_shards_per_hour": 78.21535807894338,
     "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": 1.209377144001337,
-      "bytes": 5838608,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.249938580000162,
       "bytes": 4790943,
@@ -562,6 +556,12 @@
       "bytes": 7585902,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2029692989999603,
+      "bytes": 1075789,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 298400, "loss": 0.6268672812730074, "lr": 0.0003, "grad_norm": 0.14880373418193077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43677955154, "tokens_per_second": 149101.95634820187, "data_wait_s": 1.704800160950981e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 298500, "loss": 0.5878512840718031, "lr": 0.0003, "grad_norm": 0.16763574281207322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43692508028, "tokens_per_second": 149412.37226701714, "data_wait_s": 1.4777997421333566e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 298600, "loss": 0.5420547910034657, "lr": 0.0003, "grad_norm": 0.13720762271914558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43707297334, "tokens_per_second": 150862.84949232306, "data_wait_s": 6.733000191161409e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 298700, "loss": 0.5739942295476794, "lr": 0.0003, "grad_norm": 0.1355857544130264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43721879925, "tokens_per_second": 149882.62072670835, "data_wait_s": 1.0290001227986068e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df3c7f00a2b6bd433623ddc18eecac414ef2b67f797f798851cf7b8883e40ffe
-size 392226
+oid sha256:c61eedb7664f71ed806f3fde75b74fc608455d04c8e20fc81cff036694470a6f
+size 394344