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": 74340.84463611433,
-      "required_pages_per_second_upper_estimate": 0.006443675997910235
+      "required_token_positions_per_second": 74601.17510783063,
+      "required_pages_per_second_upper_estimate": 0.006466240783397055
     },
     {
       "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": 9292.605579514291,
-      "required_pages_per_second_upper_estimate": 0.0037069593025029087
+      "required_token_positions_per_second": 9325.146888478828,
+      "required_pages_per_second_upper_estimate": 0.0037199405171847886
     },
     {
       "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": 18585.211159028582,
-      "required_pages_per_second_upper_estimate": 0.004299361513096325
+      "required_token_positions_per_second": 18650.293776957657,
+      "required_pages_per_second_upper_estimate": 0.004314417231339261
     },
     {
       "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": 27877.81673854287,
-      "required_pages_per_second_upper_estimate": 0.0016976127757766803
+      "required_token_positions_per_second": 27975.440665436483,
+      "required_pages_per_second_upper_estimate": 0.0017035575607313412
     },
     {
       "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": 27877.81673854287,
-      "required_pages_per_second_upper_estimate": 0.002146111503268909
+      "required_token_positions_per_second": 27975.440665436483,
+      "required_pages_per_second_upper_estimate": 0.0021536268634015053
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27877.81673854287,
-      "required_pages_per_second_upper_estimate": 0.3461491828419592
+      "required_token_positions_per_second": 27975.440665436483,
+      "required_pages_per_second_upper_estimate": 0.3473613452877123
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06771923039064376,
-    "producer_p95_seconds": 0.18572605600002134,
-    "gpu_update_seconds": 0.9917562863441478,
-    "producer_batches_per_second": 14.766854174677128,
-    "gpu_batches_per_second": 1.0083122373604916,
-    "producer_headroom": 14.645120457263364,
+    "producer_batch_seconds": 0.08184817453123827,
+    "producer_p95_seconds": 0.18137697000020125,
+    "gpu_update_seconds": 0.9882954242132445,
+    "producer_batches_per_second": 12.217743471094018,
+    "gpu_batches_per_second": 1.0118431953644562,
+    "producer_headroom": 12.074739966693459,
     "sustainable": true,
-    "buffer_seconds": 60.49713346699301,
+    "buffer_seconds": 60.286020877007914,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493985.76548538153,
-    "upload_bytes_per_second": 5744109.02581228,
+    "generation_bytes_per_second": 517114.51073620556,
+    "upload_bytes_per_second": 5758205.109134534,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.99784587308011,
+    "estimated_full_shards_per_hour": 110.96073619427324,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 1150905,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 0.9866051019998849,
+      "bytes": 6307267,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 86700, "loss": 0.6089567504823208, "lr": 0.0003, "grad_norm": 0.1368707770063171, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12687942064, "tokens_per_second": 146885.27042011803, "data_wait_s": 1.192699983221246e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86800, "loss": 0.6323569770902395, "lr": 0.0003, "grad_norm": 0.13840676431523363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12702535251, "tokens_per_second": 148607.16218508166, "data_wait_s": 8.262000847025774e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 86900, "loss": 0.5962490197271109, "lr": 0.0003, "grad_norm": 0.1517551616237985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12717146045, "tokens_per_second": 147696.89799948194, "data_wait_s": 1.0507999832043424e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 87000, "loss": 0.6176202818751335, "lr": 0.0003, "grad_norm": 0.12908852094214038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12731783694, "tokens_per_second": 148367.29145016827, "data_wait_s": 1.1268999514868483e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8b311a898034076995ec0f45f77dbe8fe544ae7a2f8cd8509ac1c4c94a57984
-size 106848
+oid sha256:c6b02b0ce0392c69ebfef11546ccdba16eb29c72c2a652087675038c8773da7f
+size 108966