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": 77537.2991734011,
-      "required_pages_per_second_upper_estimate": 0.005364210268136913
+      "required_token_positions_per_second": 77374.2199352696,
+      "required_pages_per_second_upper_estimate": 0.0053529280680470115
     },
     {
       "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": 9692.162396675138,
-      "required_pages_per_second_upper_estimate": 0.004057906000121893
+      "required_token_positions_per_second": 9671.7774919087,
+      "required_pages_per_second_upper_estimate": 0.004049371266181404
     },
     {
       "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": 19384.324793350275,
-      "required_pages_per_second_upper_estimate": 0.004037243655433416
+      "required_token_positions_per_second": 19343.5549838174,
+      "required_pages_per_second_upper_estimate": 0.004028752379279898
     },
     {
       "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": 29076.48719002541,
-      "required_pages_per_second_upper_estimate": 0.001743091009816767
+      "required_token_positions_per_second": 29015.3324757261,
+      "required_pages_per_second_upper_estimate": 0.0017394248780723653
     },
     {
       "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": 29076.48719002541,
-      "required_pages_per_second_upper_estimate": 0.0022605689850735762
+      "required_token_positions_per_second": 29015.3324757261,
+      "required_pages_per_second_upper_estimate": 0.002255814475027965
     },
     {
       "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": 29076.48719002541,
-      "required_pages_per_second_upper_estimate": 0.37296195777408464
+      "required_token_positions_per_second": 29015.3324757261,
+      "required_pages_per_second_upper_estimate": 0.3721775307618694
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.054204198937519266,
-    "producer_p95_seconds": 0.1608956360000775,
-    "gpu_update_seconds": 0.950871397198371,
-    "producer_batches_per_second": 18.448755255154527,
-    "gpu_batches_per_second": 1.0516669267225627,
-    "producer_headroom": 17.542393686039574,
+    "producer_batch_seconds": 0.04917161866406161,
+    "producer_p95_seconds": 0.1382736109999314,
+    "gpu_update_seconds": 0.9528755192838133,
+    "producer_batches_per_second": 20.33693474343314,
+    "gpu_batches_per_second": 1.0494550229935655,
+    "producer_headroom": 19.378567254289877,
     "sustainable": true,
-    "buffer_seconds": 60.855769420695744,
+    "buffer_seconds": 60.03115771488024,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -23,3 +23,4 @@
 {"step": 687300, "loss": 0.5587503910064697, "lr": 8e-05, "grad_norm": 0.20992206007327197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100602552012, "tokens_per_second": 152279.58006398738, "data_wait_s": 1.1000000085914508e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687400, "loss": 0.5115691423416138, "lr": 8e-05, "grad_norm": 0.19852005288075664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100617037193, "tokens_per_second": 151616.1899802224, "data_wait_s": 1.4263000139180804e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687500, "loss": 0.5395483374595642, "lr": 8e-05, "grad_norm": 0.19677089833881958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100631644735, "tokens_per_second": 153012.2355017569, "data_wait_s": 1.9391000023460947e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687600, "loss": 0.5459145903587341, "lr": 8e-05, "grad_norm": 0.2509706096515456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100646213173, "tokens_per_second": 152792.07975177924, "data_wait_s": 1.332200008619111e-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 sha256:11715c0be93a61f04bbded6f972d2dd300bdc2a6911b881a7743be2ed99d8d54
-size 53115
+oid sha256:5fbdf1632728ee562f3582b905ba2b07df56ddbee43b8d224a56c267cefdc0f1
+size 55233