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": 77340.18956064493,
-      "required_pages_per_second_upper_estimate": 0.005350573767769128
+      "required_token_positions_per_second": 77207.56510669584,
+      "required_pages_per_second_upper_estimate": 0.0053413985003138
     },
     {
       "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": 9667.523695080616,
-      "required_pages_per_second_upper_estimate": 0.004047590290278864
+      "required_token_positions_per_second": 9650.94563833698,
+      "required_pages_per_second_upper_estimate": 0.0040406494041094945
     },
     {
       "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": 19335.04739016123,
-      "required_pages_per_second_upper_estimate": 0.00402698047188165
+      "required_token_positions_per_second": 19301.89127667396,
+      "required_pages_per_second_upper_estimate": 0.0040200749278277645
     },
     {
       "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": 29002.57108524185,
-      "required_pages_per_second_upper_estimate": 0.0017386598521983495
+      "required_token_positions_per_second": 28952.83691501094,
+      "required_pages_per_second_upper_estimate": 0.0017356783646326874
     },
     {
       "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": 29002.57108524185,
-      "required_pages_per_second_upper_estimate": 0.0022548223330492393
+      "required_token_positions_per_second": 28952.83691501094,
+      "required_pages_per_second_upper_estimate": 0.002250955720071282
     },
     {
       "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": 29002.57108524185,
-      "required_pages_per_second_upper_estimate": 0.37201384134685095
+      "required_token_positions_per_second": 28952.83691501094,
+      "required_pages_per_second_upper_estimate": 0.3713759048115204
     }
   ],
   "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": 0.09251139235945516,
-    "producer_p95_seconds": 0.2546899970002414,
-    "gpu_update_seconds": 0.9532947930285005,
-    "producer_batches_per_second": 10.809479508366676,
-    "gpu_batches_per_second": 1.0489934564974626,
-    "producer_headroom": 10.304620530674228,
+    "producer_batch_seconds": 0.10594242977316526,
+    "producer_p95_seconds": 0.24617448299977696,
+    "gpu_update_seconds": 0.9549323294694334,
+    "producer_batches_per_second": 9.439088778132739,
+    "gpu_batches_per_second": 1.047194622215384,
+    "producer_headroom": 9.013691034971085,
     "sustainable": true,
-    "buffer_seconds": 60.057571960795535,
+    "buffer_seconds": 60.1607367565743,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 517672.98346856836,
-    "upload_bytes_per_second": 4886559.329300388,
+    "generation_bytes_per_second": 514677.80507525976,
+    "upload_bytes_per_second": 4888334.023617151,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.08057144205846,
+    "estimated_full_shards_per_hour": 110.43787588303894,
     "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": 1.1600040840003203,
-      "bytes": 6119662,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8502458040002239,
       "bytes": 891680,
@@ -562,6 +556,12 @@
       "bytes": 6818785,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1884660100004112,
+      "bytes": 6414581,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 907500, "loss": 0.5206209421157837, "lr": 3e-05, "grad_norm": 0.32198816843496814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132831872870, "tokens_per_second": 152057.82358131764, "data_wait_s": 1.4343000657390803e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907600, "loss": 0.5512479543685913, "lr": 3e-05, "grad_norm": 0.3486921869976473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132846430728, "tokens_per_second": 152014.17575299123, "data_wait_s": 1.1598996934480965e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907700, "loss": 0.524348795413971, "lr": 3e-05, "grad_norm": 0.32118624783546756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132861030216, "tokens_per_second": 152002.71050669058, "data_wait_s": 1.8541999452281743e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 907800, "loss": 0.5214112997055054, "lr": 3e-05, "grad_norm": 0.31540677744893886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132875511591, "tokens_per_second": 150823.5107633387, "data_wait_s": 1.4773999282624573e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ad35c6edd236284512a0978242605538cb4a449e8917c6823991fbc61403ec2
-size 484083
+oid sha256:a991aaa350396670ee0b73b81037dfdf8cb1cd952ef1bded45acdd0a93d48f95
+size 486201