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": 74941.04743367953,
-      "required_pages_per_second_upper_estimate": 0.005184595548648975
+      "required_token_positions_per_second": 76141.05339350706,
+      "required_pages_per_second_upper_estimate": 0.005267614745347285
     },
     {
       "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": 9367.630929209941,
-      "required_pages_per_second_upper_estimate": 0.003922031451681894
+      "required_token_positions_per_second": 9517.631674188382,
+      "required_pages_per_second_upper_estimate": 0.0039848336312326175
     },
     {
       "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": 18735.261858419883,
-      "required_pages_per_second_upper_estimate": 0.003902060963028074
+      "required_token_positions_per_second": 19035.263348376764,
+      "required_pages_per_second_upper_estimate": 0.003964543361814773
     },
     {
       "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": 28102.892787629822,
-      "required_pages_per_second_upper_estimate": 0.0016847255119857274
+      "required_token_positions_per_second": 28552.895022565146,
+      "required_pages_per_second_upper_estimate": 0.0017117024588564724
     },
     {
       "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": 28102.892787629822,
-      "required_pages_per_second_upper_estimate": 0.0021848763026765185
+      "required_token_positions_per_second": 28552.895022565146,
+      "required_pages_per_second_upper_estimate": 0.0022198619970920925
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28102.892787629822,
-      "required_pages_per_second_upper_estimate": 0.3604737341443776
+      "required_token_positions_per_second": 28552.895022565146,
+      "required_pages_per_second_upper_estimate": 0.36624587963937283
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.276267543921886,
-    "producer_p95_seconds": 0.17134041400004207,
-    "gpu_update_seconds": 0.9838133109261245,
-    "producer_batches_per_second": 3.6196796257860373,
-    "gpu_batches_per_second": 1.0164530088118426,
-    "producer_headroom": 3.5610889971363964,
+    "producer_batch_seconds": 0.10098976439062568,
+    "producer_p95_seconds": 0.1695617619998302,
+    "gpu_update_seconds": 0.968308116502722,
+    "producer_batches_per_second": 9.901993593449996,
+    "gpu_batches_per_second": 1.0327291313138434,
+    "producer_headroom": 9.588180766095585,
     "sustainable": true,
-    "buffer_seconds": 60.012611966493594,
+    "buffer_seconds": 60.03510322316876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 527787.7338910318,
-    "upload_bytes_per_second": 4113905.8225012827,
+    "generation_bytes_per_second": 529303.2397908698,
+    "upload_bytes_per_second": 4204607.146147268,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.25096142338005,
+    "estimated_full_shards_per_hour": 113.5761537103135,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 6196201,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.8011739470002794,
+      "bytes": 11923984,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 846200, "loss": 0.5164526104927063, "lr": 3e-05, "grad_norm": 0.31853155076772827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123860162102, "tokens_per_second": 152481.32985085127, "data_wait_s": 1.1221000022487715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 846300, "loss": 0.5566064119338989, "lr": 3e-05, "grad_norm": 0.3027356994563249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123874783257, "tokens_per_second": 150932.73232555485, "data_wait_s": 1.8663000446395017e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 846400, "loss": 0.5073434710502625, "lr": 3e-05, "grad_norm": 0.3002456759661649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123889310403, "tokens_per_second": 150441.5557627335, "data_wait_s": 7.152999387471937e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 846500, "loss": 0.5031244158744812, "lr": 3e-05, "grad_norm": 0.28552692825837883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123903826351, "tokens_per_second": 150534.72185948706, "data_wait_s": 1.5296000128728338e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02305c8d855c12be30d8fcb1608286667a0bc24ebf3626727bf5d8d6be548d8a
-size 83550
+oid sha256:84d192582913e5f7a8654e00e4d5a41d38032f5e7588641376a91d392cfcb59c
+size 85668