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": 76352.10388818971,
-      "required_pages_per_second_upper_estimate": 0.005282215708274047
+      "required_token_positions_per_second": 76833.61684008052,
+      "required_pages_per_second_upper_estimate": 0.005315527891549831
     },
     {
       "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": 9544.012986023714,
-      "required_pages_per_second_upper_estimate": 0.003995878935593634
+      "required_token_positions_per_second": 9604.202105010065,
+      "required_pages_per_second_upper_estimate": 0.004021078862821489
     },
     {
       "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": 19088.025972047428,
-      "required_pages_per_second_upper_estimate": 0.003975532424881415
+      "required_token_positions_per_second": 19208.40421002013,
+      "required_pages_per_second_upper_estimate": 0.004000604037263511
     },
     {
       "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": 28632.03895807114,
-      "required_pages_per_second_upper_estimate": 0.0017164470169442646
+      "required_token_positions_per_second": 28812.606315030196,
+      "required_pages_per_second_upper_estimate": 0.00172727175428357
     },
     {
       "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": 28632.03895807114,
-      "required_pages_per_second_upper_estimate": 0.002226015089960302
+      "required_token_positions_per_second": 28812.606315030196,
+      "required_pages_per_second_upper_estimate": 0.002240053407732002
     },
     {
       "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": 28632.03895807114,
-      "required_pages_per_second_upper_estimate": 0.36726105306590656
+      "required_token_positions_per_second": 28812.606315030196,
+      "required_pages_per_second_upper_estimate": 0.3695771772428547
     }
   ],
   "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": 0.15200320324987615,
-    "producer_p95_seconds": 0.25585103900084505,
-    "gpu_update_seconds": 0.9656315444557696,
-    "producer_batches_per_second": 6.578808726524747,
-    "gpu_batches_per_second": 1.0355916868515314,
-    "producer_headroom": 6.352705231273186,
+    "producer_batch_seconds": 0.10449727107803142,
+    "producer_p95_seconds": 0.25552427600268857,
+    "gpu_update_seconds": 0.9595799733527517,
+    "producer_batches_per_second": 9.569627892514708,
+    "gpu_batches_per_second": 1.0421226242415436,
+    "producer_headroom": 9.182823278095013,
     "sustainable": true,
-    "buffer_seconds": 60.83478730071349,
+    "buffer_seconds": 60.453538321223355,
     "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": 318849.02154025715,
-    "upload_bytes_per_second": 4989684.175797677,
+    "generation_bytes_per_second": 319514.78736261826,
+    "upload_bytes_per_second": 4862982.442614462,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.41757759719644,
+    "estimated_full_shards_per_hour": 68.56043544443999,
     "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": 1.171707494999282,
-      "bytes": 6154958,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3786574580008164,
       "bytes": 2401890,
@@ -562,6 +556,12 @@
       "bytes": 15647853,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 2.7048667290000594,
+      "bytes": 2731260,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 785800, "loss": 0.4968486726284027, "lr": 3e-05, "grad_norm": 0.25902219601183707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115018910117, "tokens_per_second": 152449.77224550906, "data_wait_s": 9.759001841302961e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 785900, "loss": 0.5158545970916748, "lr": 3e-05, "grad_norm": 0.26559567817473356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115033482450, "tokens_per_second": 150697.70523083094, "data_wait_s": 8.023998816497624e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 786000, "loss": 0.5087425112724304, "lr": 3e-05, "grad_norm": 0.25850710350293415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115048196311, "tokens_per_second": 152991.88990924697, "data_wait_s": 1.1832999007310718e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 786100, "loss": 0.5528032183647156, "lr": 3e-05, "grad_norm": 0.28188182765323166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115062769846, "tokens_per_second": 150874.42046029482, "data_wait_s": 1.3863998901797459e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9b44dae4fbfd83ade8563351fd9bc0f063761543bc748afbc75537f1517c756
-size 554760
+oid sha256:39b8168d42b48bf5fb6ff1a2339742c6ee595941797852285a5464b278a39864
+size 556878