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": 76510.94928826316,
-      "required_pages_per_second_upper_estimate": 0.006631775169078539
+      "required_token_positions_per_second": 75442.40974641254,
+      "required_pages_per_second_upper_estimate": 0.006539156869779636
     },
     {
       "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": 9563.868661032895,
-      "required_pages_per_second_upper_estimate": 0.0038151702014651728
+      "required_token_positions_per_second": 9430.301218301567,
+      "required_pages_per_second_upper_estimate": 0.003761888151548415
     },
     {
       "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": 19127.73732206579,
-      "required_pages_per_second_upper_estimate": 0.004424865392780622
+      "required_token_positions_per_second": 18860.602436603134,
+      "required_pages_per_second_upper_estimate": 0.004363068438442248
     },
     {
       "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": 28691.605983098685,
-      "required_pages_per_second_upper_estimate": 0.0017471682711479371
+      "required_token_positions_per_second": 28290.903654904705,
+      "required_pages_per_second_upper_estimate": 0.0017227676016835637
     },
     {
       "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": 28691.605983098685,
-      "required_pages_per_second_upper_estimate": 0.0022087592520276247
+      "required_token_positions_per_second": 28290.903654904705,
+      "required_pages_per_second_upper_estimate": 0.002177912077588215
     },
     {
       "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": 28691.605983098685,
-      "required_pages_per_second_upper_estimate": 0.3562537216819435
+      "required_token_positions_per_second": 28290.903654904705,
+      "required_pages_per_second_upper_estimate": 0.3512783398301986
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.5224386731014192,
-    "producer_p95_seconds": 0.35933989199838834,
-    "gpu_update_seconds": 0.9636267839550898,
-    "producer_batches_per_second": 1.9141002599665387,
-    "gpu_batches_per_second": 1.0377461654766595,
-    "producer_headroom": 1.8444782776791568,
+    "producer_batch_seconds": 0.13573866137471668,
+    "producer_p95_seconds": 0.3541406059957808,
+    "gpu_update_seconds": 0.9772752520475518,
+    "producer_batches_per_second": 7.3670978472332616,
+    "gpu_batches_per_second": 1.023253170388625,
+    "producer_headroom": 7.199682405513863,
     "sustainable": true,
-    "buffer_seconds": 60.70848738917066,
+    "buffer_seconds": 60.591065626948215,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 253571.09304862283,
-    "upload_bytes_per_second": 5679500.946338109,
+    "generation_bytes_per_second": 242383.04180104623,
+    "upload_bytes_per_second": 5729717.121183149,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.410453735294475,
+    "estimated_full_shards_per_hour": 52.009758382068064,
     "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": 0.9228753829993366,
-      "bytes": 2415467,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.343091220001952,
       "bytes": 11741900,
@@ -562,6 +556,12 @@
       "bytes": 11555501,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.030656152004667,
+      "bytes": 6775920,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -275,3 +275,4 @@
 {"step": 307700, "loss": 0.5667116083204746, "lr": 0.0003, "grad_norm": 0.13840983292867118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45039123260, "tokens_per_second": 149256.11065812217, "data_wait_s": 1.0973999451380223e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 307800, "loss": 0.5906594209372997, "lr": 0.0003, "grad_norm": 0.14395073653544735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45053581907, "tokens_per_second": 147806.31520957616, "data_wait_s": 1.3195000065024942e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 307900, "loss": 0.5711268819868565, "lr": 0.0003, "grad_norm": 0.16283465958012144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45068149169, "tokens_per_second": 118495.73769390934, "data_wait_s": 9.321003744844347e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 308000, "loss": 0.5821310188621283, "lr": 0.0003, "grad_norm": 0.1793600506390255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45082883292, "tokens_per_second": 150587.10168041612, "data_wait_s": 1.3660996046382934e-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 sha256:0f93e384aa25d7b1c646b93f735ad608ef1e03b6272d9f25cfb3e65ec94537d5
-size 590766
+oid sha256:00c6269d3bc74178c5893995816ff4001bd3916feff73944fa25cd58227fc52b
+size 592884