Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75483.18640121476,
-      "required_pages_per_second_upper_estimate": 0.006542691286870406
+      "required_token_positions_per_second": 75702.33370734517,
+      "required_pages_per_second_upper_estimate": 0.00656168641994202
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9435.398300151845,
-      "required_pages_per_second_upper_estimate": 0.0037639214537066555
+      "required_token_positions_per_second": 9462.791713418146,
+      "required_pages_per_second_upper_estimate": 0.00377484909582661
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18870.79660030369,
-      "required_pages_per_second_upper_estimate": 0.004365426678803218
+      "required_token_positions_per_second": 18925.583426836292,
+      "required_pages_per_second_upper_estimate": 0.00437810064690632
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28306.194900455535,
-      "required_pages_per_second_upper_estimate": 0.0017236987583106442
+      "required_token_positions_per_second": 28388.37514025444,
+      "required_pages_per_second_upper_estimate": 0.0017287031037480016
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28306.194900455535,
-      "required_pages_per_second_upper_estimate": 0.002179089239999595
+      "required_token_positions_per_second": 28388.37514025444,
+      "required_pages_per_second_upper_estimate": 0.002185415702348778
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28306.194900455535,
-      "required_pages_per_second_upper_estimate": 0.35146820592343314
+      "required_token_positions_per_second": 28388.37514025444,
+      "required_pages_per_second_upper_estimate": 0.3524886094621657
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07478907846098082,
-    "producer_p95_seconds": 0.20420235600067826,
-    "gpu_update_seconds": 0.9767473197026231,
-    "producer_batches_per_second": 13.370936246015159,
-    "gpu_batches_per_second": 1.0238062391657818,
-    "producer_headroom": 13.060026140209958,
+    "producer_batch_seconds": 0.05522118604679349,
+    "producer_p95_seconds": 0.19276577699929476,
+    "gpu_update_seconds": 0.973919777493549,
+    "producer_batches_per_second": 18.10899170388367,
+    "gpu_batches_per_second": 1.026778614737212,
+    "producer_headroom": 17.63670517087891,
     "sustainable": true,
-    "buffer_seconds": 60.55833382156263,
+    "buffer_seconds": 60.383026204600036,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345152.5905722208,
-    "upload_bytes_per_second": 6062939.340143687,
+    "generation_bytes_per_second": 344723.44766883296,
+    "upload_bytes_per_second": 6114155.265965811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.06171119570702,
+    "estimated_full_shards_per_hour": 73.96962711857549,
     "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": 0.7539118750000853,
-      "bytes": 789098,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 0.9344837560001906,
       "bytes": 6249982,
@@ -562,6 +556,12 @@
       "bytes": 6606039,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.092060394001237,
+      "bytes": 6397904,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 50600, "loss": 0.6595828458666801, "lr": 0.0003, "grad_norm": 0.14864370134429494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7405328376, "tokens_per_second": 151175.29936112068, "data_wait_s": 8.854000043356791e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 50700, "loss": 0.6321609169244766, "lr": 0.0003, "grad_norm": 0.12893147333020086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7419983847, "tokens_per_second": 148425.897060511, "data_wait_s": 9.569001122144982e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 50800, "loss": 0.6277779079973698, "lr": 0.0003, "grad_norm": 0.13406036670035498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7434572545, "tokens_per_second": 149314.30217967523, "data_wait_s": 8.113000149023719e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 50900, "loss": 0.617812218144536, "lr": 0.0003, "grad_norm": 0.1325574122791432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7449321774, "tokens_per_second": 151055.9601549752, "data_wait_s": 7.963999451021664e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01e4016031c5f0d267fcb0a568909dfedb3af5f12a9dda59fc1daf1a813f0cf3
-size 230475
+oid sha256:3e7edca66456a0bb439bf6f4c8f9f3cf85944dec6b32c8d51f856d875e3b2d98
+size 232593