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": 76712.68508891354,
-      "required_pages_per_second_upper_estimate": 0.005307161552403777
+      "required_token_positions_per_second": 77076.51107097555,
+      "required_pages_per_second_upper_estimate": 0.005332331878035413
     },
     {
       "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": 9589.085636114192,
-      "required_pages_per_second_upper_estimate": 0.004014749912962553
+      "required_token_positions_per_second": 9634.563883871944,
+      "required_pages_per_second_upper_estimate": 0.004033790705604918
     },
     {
       "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": 19178.171272228385,
-      "required_pages_per_second_upper_estimate": 0.003994307313617676
+      "required_token_positions_per_second": 19269.12776774389,
+      "required_pages_per_second_upper_estimate": 0.0040132511529494645
     },
     {
       "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": 28767.256908342577,
-      "required_pages_per_second_upper_estimate": 0.0017245531265971817
+      "required_token_positions_per_second": 28903.691651615834,
+      "required_pages_per_second_upper_estimate": 0.0017327321811326229
     },
     {
       "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": 28767.256908342577,
-      "required_pages_per_second_upper_estimate": 0.0022365276908329988
+      "required_token_positions_per_second": 28903.691651615834,
+      "required_pages_per_second_upper_estimate": 0.0022471348920094806
     },
     {
       "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": 28767.256908342577,
-      "required_pages_per_second_upper_estimate": 0.3689954837462651
+      "required_token_positions_per_second": 28903.691651615834,
+      "required_pages_per_second_upper_estimate": 0.37074552214076056
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06831225260944507,
-    "producer_p95_seconds": 0.23298999500184436,
-    "gpu_update_seconds": 0.9610926786690605,
-    "producer_batches_per_second": 14.63866234535116,
-    "gpu_batches_per_second": 1.0404823823908629,
-    "producer_headroom": 14.069111205625457,
+    "producer_batch_seconds": 0.5019277770938118,
+    "producer_p95_seconds": 0.24257628000123077,
+    "gpu_update_seconds": 0.9565560113652252,
+    "producer_batches_per_second": 1.9923185080332726,
+    "gpu_batches_per_second": 1.0454170880937441,
+    "producer_headroom": 1.9057642454134234,
     "sustainable": true,
-    "buffer_seconds": 60.54883875615081,
+    "buffer_seconds": 60.26302871600919,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295673.1996105879,
-    "upload_bytes_per_second": 4996132.746697853,
+    "generation_bytes_per_second": 268823.5256942857,
+    "upload_bytes_per_second": 4932462.557783562,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.44458571661213,
+    "estimated_full_shards_per_hour": 57.683270722593576,
     "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": [
-    {
-      "seconds": 0.9956274190008116,
-      "bytes": 6486340,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.1072240289995534,
       "bytes": 6495721,
@@ -562,6 +556,12 @@
       "bytes": 5788392,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0160050430022238,
+      "bytes": 1159604,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -210,3 +210,4 @@
 {"step": 461200, "loss": 0.5550450682640076, "lr": 0.0003, "grad_norm": 0.16242360347050405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67505691182, "tokens_per_second": 151115.94158695123, "data_wait_s": 1.3955999747850001e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461300, "loss": 0.541749119758606, "lr": 0.0003, "grad_norm": 0.14559382314573044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67520399023, "tokens_per_second": 152942.82485674313, "data_wait_s": 1.3496999599738047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461400, "loss": 0.5921451449394226, "lr": 0.0003, "grad_norm": 0.15366019789936122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67535009988, "tokens_per_second": 152302.3197972847, "data_wait_s": 1.3390999811235815e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 461500, "loss": 0.62558513879776, "lr": 0.0003, "grad_norm": 0.16961965440790405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67549806988, "tokens_per_second": 151838.21216204233, "data_wait_s": 1.6812999092508107e-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 sha256:1c6780a6556be04e2203c83bb5f5988602b0ef4b14f3e34e3bd21a0d72da3b0e
-size 452313
+oid sha256:983b21297af57ba7d6bb5f5d3cb535ca87ca7529c0ef5043bd9fdfaef2993c0a
+size 454431