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": 77209.24521130872,
-      "required_pages_per_second_upper_estimate": 0.0053415147338492985
+      "required_token_positions_per_second": 77064.45029573975,
+      "required_pages_per_second_upper_estimate": 0.005331497485619746
     },
     {
       "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": 9651.15565141359,
-      "required_pages_per_second_upper_estimate": 0.0040407373321991
+      "required_token_positions_per_second": 9633.056286967469,
+      "required_pages_per_second_upper_estimate": 0.004033159506263217
     },
     {
       "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": 19302.31130282718,
-      "required_pages_per_second_upper_estimate": 0.004020162408198646
+      "required_token_positions_per_second": 19266.112573934937,
+      "required_pages_per_second_upper_estimate": 0.004012623167595068
     },
     {
       "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": 28953.466954240772,
-      "required_pages_per_second_upper_estimate": 0.0017357161345225016
+      "required_token_positions_per_second": 28899.168860902406,
+      "required_pages_per_second_upper_estimate": 0.0017324610467352544
     },
     {
       "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": 28953.466954240772,
-      "required_pages_per_second_upper_estimate": 0.0022510047028501514
+      "required_token_positions_per_second": 28899.168860902406,
+      "required_pages_per_second_upper_estimate": 0.0022467832649251656
     },
     {
       "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": 28953.466954240772,
-      "required_pages_per_second_upper_estimate": 0.37138398627827723
+      "required_token_positions_per_second": 28899.168860902406,
+      "required_pages_per_second_upper_estimate": 0.370687508637683
     }
   ],
   "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": 0.0899327893751547,
-    "producer_p95_seconds": 0.22277288700206554,
-    "gpu_update_seconds": 0.9549115497531269,
-    "producer_batches_per_second": 11.11941492027451,
-    "gpu_batches_per_second": 1.0472174100926204,
-    "producer_headroom": 10.618057733867373,
+    "producer_batch_seconds": 0.37292732897654446,
+    "producer_p95_seconds": 0.2370887920005771,
+    "gpu_update_seconds": 0.9567057147240277,
+    "producer_batches_per_second": 2.6814875776049543,
+    "gpu_batches_per_second": 1.0452535033601853,
+    "producer_headroom": 2.5653944894561493,
     "sustainable": true,
-    "buffer_seconds": 60.15942763444699,
+    "buffer_seconds": 60.27246002761374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417414.99342329195,
+    "generation_bytes_per_second": 371686.07824357064,
     "upload_bytes_per_second": 4707631.099469831,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.5675406649024,
+    "estimated_full_shards_per_hour": 79.75517998199787,
     "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

@@ -174,3 +174,4 @@
 {"step": 457600, "loss": 0.5711995363235474, "lr": 0.0003, "grad_norm": 0.13963287596159765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66978781488, "tokens_per_second": 150979.49881917232, "data_wait_s": 1.1741998605430126e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457700, "loss": 0.45534321665763855, "lr": 0.0003, "grad_norm": 0.22608440001065322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66993292703, "tokens_per_second": 151032.16086570555, "data_wait_s": 1.5909001376712695e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457800, "loss": 0.5443060398101807, "lr": 0.0003, "grad_norm": 0.17473214796489223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67007858122, "tokens_per_second": 152137.92055715068, "data_wait_s": 1.589800012880005e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 457900, "loss": 0.6169613003730774, "lr": 0.0003, "grad_norm": 0.1489924384099695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67022638589, "tokens_per_second": 154146.99362011062, "data_wait_s": 8.695998985785991e-06, "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 sha256:6fb33fd489a10bdd1e510ca9cb01638c84a12257fe325981e0c95fb3125df473
-size 375282
+oid sha256:9bbb17df542e5a0293276742a65d5bc7d44ec0a4994132e22f3ff3c4a301289d
+size 377400