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": 77139.79255549624,
-      "required_pages_per_second_upper_estimate": 0.005336709837967827
+      "required_token_positions_per_second": 77091.31931848508,
+      "required_pages_per_second_upper_estimate": 0.00533335634695799
     },
     {
       "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": 9642.47406943703,
-      "required_pages_per_second_upper_estimate": 0.004037102535117561
+      "required_token_positions_per_second": 9636.414914810635,
+      "required_pages_per_second_upper_estimate": 0.004034565693605026
     },
     {
       "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": 19284.94813887406,
-      "required_pages_per_second_upper_estimate": 0.004016546119044636
+      "required_token_positions_per_second": 19272.82982962127,
+      "required_pages_per_second_upper_estimate": 0.004014022194808586
     },
     {
       "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": 28927.422208311087,
-      "required_pages_per_second_upper_estimate": 0.0017341547917720413
+      "required_token_positions_per_second": 28909.244744431904,
+      "required_pages_per_second_upper_estimate": 0.0017330650805678647
     },
     {
       "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": 28927.422208311087,
-      "required_pages_per_second_upper_estimate": 0.002248979838412849
+      "required_token_positions_per_second": 28909.244744431904,
+      "required_pages_per_second_upper_estimate": 0.0022475666205504564
     },
     {
       "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": 28927.422208311087,
-      "required_pages_per_second_upper_estimate": 0.37104991224216066
+      "required_token_positions_per_second": 28909.244744431904,
+      "required_pages_per_second_upper_estimate": 0.3708167512529586
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12584090176576979,
-    "producer_p95_seconds": 0.2816668179984845,
-    "gpu_update_seconds": 0.9557713024306915,
-    "producer_batches_per_second": 7.946541911002197,
-    "gpu_batches_per_second": 1.0462753981594,
-    "producer_headroom": 7.595076712098646,
+    "producer_batch_seconds": 0.3416976367889788,
+    "producer_p95_seconds": 0.2965564399964933,
+    "gpu_update_seconds": 0.956372269300642,
+    "producer_batches_per_second": 2.926563992064034,
+    "gpu_batches_per_second": 1.0456179378049735,
+    "producer_headroom": 2.7988846463438257,
     "sustainable": true,
-    "buffer_seconds": 60.21359205313357,
+    "buffer_seconds": 60.251452965940445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 492801.22605693585,
+    "generation_bytes_per_second": 431586.9219488519,
     "upload_bytes_per_second": 5370145.30130805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.74367128640229,
+    "estimated_full_shards_per_hour": 92.608506620876,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -269,3 +269,4 @@
 {"step": 467100, "loss": 0.5976812839508057, "lr": 0.0003, "grad_norm": 0.1841267185010909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68369643579, "tokens_per_second": 150685.24245613866, "data_wait_s": 9.10999733605422e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 467200, "loss": 0.5728641152381897, "lr": 0.0003, "grad_norm": 0.1470858756226837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68384192238, "tokens_per_second": 151337.50768905797, "data_wait_s": 2.0669001969508827e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 467300, "loss": 0.5641692280769348, "lr": 0.0003, "grad_norm": 0.16700952652729079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68398658988, "tokens_per_second": 149928.1134469061, "data_wait_s": 8.442002581432462e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 467400, "loss": 0.5775232315063477, "lr": 0.0003, "grad_norm": 0.19060062456859225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68413454837, "tokens_per_second": 153052.73668513514, "data_wait_s": 1.5683999663451687e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3c000ead56d131e1351ef8a7cfe1ce705f9e3e42f43fc4343fe21cc17d9784d
-size 578058
+oid sha256:d5ccab6ff15114a591cce304e194023dbc2c224e37f9467b67f477ae0518ea5c
+size 580176