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": 77149.58466475076,
-      "required_pages_per_second_upper_estimate": 0.0053373872788068785
+      "required_token_positions_per_second": 77023.84061816511,
+      "required_pages_per_second_upper_estimate": 0.005328688013897698
     },
     {
       "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": 9643.698083093845,
-      "required_pages_per_second_upper_estimate": 0.004037615004075358
+      "required_token_positions_per_second": 9627.980077270639,
+      "required_pages_per_second_upper_estimate": 0.004031034203266467
     },
     {
       "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": 19287.39616618769,
-      "required_pages_per_second_upper_estimate": 0.004017055978575244
+      "required_token_positions_per_second": 19255.960154541277,
+      "required_pages_per_second_upper_estimate": 0.0040105086863726725
     },
     {
       "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": 28931.094249281534,
-      "required_pages_per_second_upper_estimate": 0.0017343749250213376
+      "required_token_positions_per_second": 28883.940231811914,
+      "required_pages_per_second_upper_estimate": 0.0017315481136740496
     },
     {
       "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": 28931.094249281534,
-      "required_pages_per_second_upper_estimate": 0.0022492653234467
+      "required_token_positions_per_second": 28883.940231811914,
+      "required_pages_per_second_upper_estimate": 0.0022455993059970402
     },
     {
       "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": 28931.094249281534,
-      "required_pages_per_second_upper_estimate": 0.37109701324099914
+      "required_token_positions_per_second": 28883.940231811914,
+      "required_pages_per_second_upper_estimate": 0.3704921721349382
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06897477890623094,
-    "producer_p95_seconds": 0.157441158000438,
-    "gpu_update_seconds": 0.9556499924190252,
-    "producer_batches_per_second": 14.498052996436115,
-    "gpu_batches_per_second": 1.0464082121412592,
-    "producer_headroom": 13.855064236134798,
+    "producer_batch_seconds": 0.07648703864844464,
+    "producer_p95_seconds": 0.15648856999996497,
+    "gpu_update_seconds": 0.9572101236225836,
+    "producer_batches_per_second": 13.074110563964616,
+    "gpu_batches_per_second": 1.044702699356623,
+    "producer_headroom": 12.514670989187898,
     "sustainable": true,
-    "buffer_seconds": 60.20594952239858,
+    "buffer_seconds": 60.30423778822277,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364940.69835521246,
-    "upload_bytes_per_second": 5852080.085074493,
+    "generation_bytes_per_second": 323666.89015679655,
+    "upload_bytes_per_second": 5990126.534546178,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.30777848236352,
+    "estimated_full_shards_per_hour": 69.45138004806445,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 10375056,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.011828892000267,
+      "bytes": 8292186,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -20,3 +20,4 @@
 {"step": 487000, "loss": 0.550245463848114, "lr": 0.0003, "grad_norm": 0.15350373814620724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71280588623, "tokens_per_second": 152684.9476076727, "data_wait_s": 8.385999535676092e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 487100, "loss": 0.5828955769538879, "lr": 0.0003, "grad_norm": 0.17236955307801755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71295318846, "tokens_per_second": 153595.28166588835, "data_wait_s": 1.0404000022390392e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 487200, "loss": 0.5762869119644165, "lr": 0.0003, "grad_norm": 0.1640187829808363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71310049986, "tokens_per_second": 152882.7269540695, "data_wait_s": 1.5609999536536634e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 487300, "loss": 0.6108181476593018, "lr": 0.0003, "grad_norm": 0.1651144827919894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71324568675, "tokens_per_second": 151824.51524321374, "data_wait_s": 7.669999831705354e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2966f475eeed4ab6f74e49af974b5acfaa8988126d18b7ee1fc626a87ae78c70
-size 46761
+oid sha256:ace938269b72348b5ab801ea07a8fc333620635a9fc49a360fa0c297aac285e5
+size 48879