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": 76924.73798094835,
-      "required_pages_per_second_upper_estimate": 0.005321831863505244
+      "required_token_positions_per_second": 76363.98324610874,
+      "required_pages_per_second_upper_estimate": 0.005283037549818793
     },
     {
       "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": 9615.592247618544,
-      "required_pages_per_second_upper_estimate": 0.004025847677678434
+      "required_token_positions_per_second": 9545.497905763592,
+      "required_pages_per_second_upper_estimate": 0.003996500640480071
     },
     {
       "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": 19231.18449523709,
-      "required_pages_per_second_upper_estimate": 0.004005348569917684
+      "required_token_positions_per_second": 19090.995811527184,
+      "required_pages_per_second_upper_estimate": 0.003976150964125114
     },
     {
       "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": 28846.776742855633,
-      "required_pages_per_second_upper_estimate": 0.0017293202192564827
+      "required_token_positions_per_second": 28636.493717290778,
+      "required_pages_per_second_upper_estimate": 0.0017167140729574583
     },
     {
       "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": 28846.776742855633,
-      "required_pages_per_second_upper_estimate": 0.002242710008195602
+      "required_token_positions_per_second": 28636.493717290778,
+      "required_pages_per_second_upper_estimate": 0.0022263614278952132
     },
     {
       "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": 28846.776742855633,
-      "required_pages_per_second_upper_estimate": 0.37001547880165253
+      "required_token_positions_per_second": 28636.493717290778,
+      "required_pages_per_second_upper_estimate": 0.36731819393402826
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.050336903656251764,
-    "producer_p95_seconds": 0.17160112200053845,
-    "gpu_update_seconds": 0.9584433036126808,
-    "producer_batches_per_second": 19.866140492648313,
-    "gpu_batches_per_second": 1.043358533812776,
-    "producer_headroom": 19.0405693238075,
+    "producer_batch_seconds": 0.07321511903123934,
+    "producer_p95_seconds": 0.17565104400000564,
+    "gpu_update_seconds": 0.9654813285785081,
+    "producer_batches_per_second": 13.658381127172943,
+    "gpu_batches_per_second": 1.035752810955251,
+    "producer_headroom": 13.186911956894555,
     "sustainable": true,
-    "buffer_seconds": 60.38192812759889,
+    "buffer_seconds": 60.82532370044601,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308240.8461012944,
-    "upload_bytes_per_second": 5217920.8349596895,
+    "generation_bytes_per_second": 306871.45631953445,
+    "upload_bytes_per_second": 5227695.851804509,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.1413100936806,
+    "estimated_full_shards_per_hour": 65.84747092427754,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 11874214,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0771185140001762,
+      "bytes": 6044727,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -52,3 +52,4 @@
 {"step": 890200, "loss": 0.5247633457183838, "lr": 3e-05, "grad_norm": 0.3070394859758585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130300368156, "tokens_per_second": 150711.66640688496, "data_wait_s": 1.783500010787975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890300, "loss": 0.5213205218315125, "lr": 3e-05, "grad_norm": 0.3233589255494616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130315114383, "tokens_per_second": 154146.30505007337, "data_wait_s": 7.2249995355377905e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 890400, "loss": 0.5553457736968994, "lr": 3e-05, "grad_norm": 0.32650330970892855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130329770014, "tokens_per_second": 152760.38046780383, "data_wait_s": 1.0931999895547051e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 890500, "loss": 0.5742422938346863, "lr": 3e-05, "grad_norm": 0.33076722199345765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130344441750, "tokens_per_second": 153408.29754627956, "data_wait_s": 1.5789999451953918e-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 sha256:4e6b1e4aba6ef2c3b1ca15a9e1258b9ffe037c52230d04dd46da566f5dd075d1
-size 115320
+oid sha256:cfca4913f1c5e8898585f8f5e3e32dbeb2e0b1d4f73eabcc706a7f8dd19c149e
+size 117438