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": 77268.00806920457,
-      "required_pages_per_second_upper_estimate": 0.005345580084707149
+      "required_token_positions_per_second": 76842.87717471927,
+      "required_pages_per_second_upper_estimate": 0.005316168542987076
     },
     {
       "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": 9658.501008650572,
-      "required_pages_per_second_upper_estimate": 0.004043812679885722
+      "required_token_positions_per_second": 9605.359646839908,
+      "required_pages_per_second_upper_estimate": 0.004021563501413423
     },
     {
       "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": 19317.002017301143,
-      "required_pages_per_second_upper_estimate": 0.004023222096603379
+      "required_token_positions_per_second": 19210.719293679816,
+      "required_pages_per_second_upper_estimate": 0.004001086208136963
     },
     {
       "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": 28975.50302595171,
-      "required_pages_per_second_upper_estimate": 0.001737037163374439
+      "required_token_positions_per_second": 28816.07894051973,
+      "required_pages_per_second_upper_estimate": 0.0017274799328792752
     },
     {
       "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": 28975.50302595171,
-      "required_pages_per_second_upper_estimate": 0.002252717910499241
+      "required_token_positions_per_second": 28816.07894051973,
+      "required_pages_per_second_upper_estimate": 0.0022403233890893478
     },
     {
       "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": 28975.50302595171,
-      "required_pages_per_second_upper_estimate": 0.37166664134569477
+      "required_token_positions_per_second": 28816.07894051973,
+      "required_pages_per_second_upper_estimate": 0.3696217203540197
     }
   ],
   "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": 0.10184826949218717,
-    "producer_p95_seconds": 0.17568903199935448,
-    "gpu_update_seconds": 0.954185332873678,
-    "producer_batches_per_second": 9.818527157957362,
-    "gpu_batches_per_second": 1.0480144323622582,
-    "producer_headroom": 9.368694604544793,
+    "producer_batch_seconds": 0.09810352032030067,
+    "producer_p95_seconds": 0.18792460300028324,
+    "gpu_update_seconds": 0.9594643343762765,
+    "producer_batches_per_second": 10.19331413118586,
+    "gpu_batches_per_second": 1.0422482255685666,
+    "producer_headroom": 9.780121357966534,
     "sustainable": true,
-    "buffer_seconds": 60.113675971041715,
+    "buffer_seconds": 60.44625306570542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481798.8063817813,
-    "upload_bytes_per_second": 5132989.268753572,
+    "generation_bytes_per_second": 501816.7979198709,
+    "upload_bytes_per_second": 1896141.3924091377,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.38280814733581,
+    "estimated_full_shards_per_hour": 107.67820313641639,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 5957600,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 25.62068473599993,
+      "bytes": 7109407,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 690400, "loss": 0.5102057456970215, "lr": 8e-05, "grad_norm": 0.20372598556241328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101055634316, "tokens_per_second": 152507.45737045284, "data_wait_s": 1.594600053067552e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690500, "loss": 0.5640904903411865, "lr": 8e-05, "grad_norm": 0.22608068633271008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101070357154, "tokens_per_second": 154020.0132578584, "data_wait_s": 1.705900012893835e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690600, "loss": 0.5559753179550171, "lr": 8e-05, "grad_norm": 0.18827603670183796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101085098989, "tokens_per_second": 153872.65503392747, "data_wait_s": 1.7520999790576752e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 690700, "loss": 0.5771751403808594, "lr": 8e-05, "grad_norm": 0.20993416394020747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101099755045, "tokens_per_second": 152782.80627288757, "data_wait_s": 1.4968999494158197e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fe43a432ed6fb4625e5925f2cb12a9b86487066664db79e07ab0f25ecae31a3
-size 119556
+oid sha256:9bd89cbb03bc4d2322db4c0ac4a965fcd984bba95576ad503f9e672ed245e735
+size 121674