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": 77247.77051665467,
-      "required_pages_per_second_upper_estimate": 0.005344180003864155
+      "required_token_positions_per_second": 76886.0870132381,
+      "required_pages_per_second_upper_estimate": 0.005319157899876447
     },
     {
       "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": 9655.971314581833,
-      "required_pages_per_second_upper_estimate": 0.00404275354980516
+      "required_token_positions_per_second": 9610.760876654762,
+      "required_pages_per_second_upper_estimate": 0.004023824883546397
     },
     {
       "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": 19311.942629163666,
-      "required_pages_per_second_upper_estimate": 0.00402216835947938
+      "required_token_positions_per_second": 19221.521753309524,
+      "required_pages_per_second_upper_estimate": 0.004003336075597812
     },
     {
       "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": 28967.9139437455,
-      "required_pages_per_second_upper_estimate": 0.0017365822094840345
+      "required_token_positions_per_second": 28832.282629964287,
+      "required_pages_per_second_upper_estimate": 0.0017284513193198754
     },
     {
       "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": 28967.9139437455,
-      "required_pages_per_second_upper_estimate": 0.0022521278927385533
+      "required_token_positions_per_second": 28832.282629964287,
+      "required_pages_per_second_upper_estimate": 0.002241583154671165
     },
     {
       "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": 28967.9139437455,
-      "required_pages_per_second_upper_estimate": 0.37156929674767514
+      "required_token_positions_per_second": 28832.282629964287,
+      "required_pages_per_second_upper_estimate": 0.36982956388404825
     }
   ],
   "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": 0.06336448393750427,
-    "producer_p95_seconds": 0.12817057600000226,
-    "gpu_update_seconds": 0.9544353125907264,
-    "producer_batches_per_second": 15.781711423489057,
-    "gpu_batches_per_second": 1.0477399429884802,
-    "producer_headroom": 15.062622675694417,
+    "producer_batch_seconds": 0.0692479436562552,
+    "producer_p95_seconds": 0.13218428600009702,
+    "gpu_update_seconds": 0.958925117197155,
+    "producer_batches_per_second": 14.440862027094571,
+    "gpu_batches_per_second": 1.0428342965120185,
+    "producer_headroom": 13.847705311759606,
     "sustainable": true,
-    "buffer_seconds": 60.12942469321577,
+    "buffer_seconds": 60.41228238342077,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 55,
+    "upload_pages_per_commit": 58,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374353.52534401446,
-    "upload_bytes_per_second": 3622099.0301886583,
+    "generation_bytes_per_second": 278534.8209568656,
+    "upload_bytes_per_second": 4001915.8121096306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.32755203476262,
+    "estimated_full_shards_per_hour": 59.76708861855961,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 6002577,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2785650519999763,
+      "bytes": 7789202,
+      "pages": 19,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -4,3 +4,4 @@
 {"step": 604800, "loss": 0.551197350025177, "lr": 0.0001, "grad_norm": 0.14659759872408634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88524662114, "tokens_per_second": 138614.38636559332, "data_wait_s": 1.7399000057594094e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 604900, "loss": 0.5269398093223572, "lr": 0.0001, "grad_norm": 0.14970845497430949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88539299377, "tokens_per_second": 151637.69207242082, "data_wait_s": 8.543000035388104e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 605000, "loss": 0.5390992760658264, "lr": 0.0001, "grad_norm": 0.14674679412242164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88553910942, "tokens_per_second": 129367.78419255553, "data_wait_s": 1.1035000170522835e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 605100, "loss": 0.5622690320014954, "lr": 0.0001, "grad_norm": 0.14328663181039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88568593606, "tokens_per_second": 139609.19308701303, "data_wait_s": 1.6875000028448994e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:241415ea1c589a80401ab37cf6489785c002a082da7aee7f181c97e6f4c2c2be
-size 13656
+oid sha256:96d4fd31f2a6159a95340fd2fd4f4eb265a886b5eb9b323a4cbd9a77b678205d
+size 15774