Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75606.51235961333,
-      "required_pages_per_second_upper_estimate": 0.00655338086837761
+      "required_token_positions_per_second": 75734.50043516197,
+      "required_pages_per_second_upper_estimate": 0.006564474550383355
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9450.814044951667,
-      "required_pages_per_second_upper_estimate": 0.003770071024793229
+      "required_token_positions_per_second": 9466.812554395246,
+      "required_pages_per_second_upper_estimate": 0.0037764530694093055
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18901.628089903334,
-      "required_pages_per_second_upper_estimate": 0.004372559001306411
+      "required_token_positions_per_second": 18933.625108790493,
+      "required_pages_per_second_upper_estimate": 0.004379960948497657
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28352.442134855002,
-      "required_pages_per_second_upper_estimate": 0.001726514972245085
+      "required_token_positions_per_second": 28400.437663185738,
+      "required_pages_per_second_upper_estimate": 0.0017294376481073507
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28352.442134855002,
-      "required_pages_per_second_upper_estimate": 0.0021826494801242043
+      "required_token_positions_per_second": 28400.437663185738,
+      "required_pages_per_second_upper_estimate": 0.00218634430875523
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28352.442134855002,
-      "required_pages_per_second_upper_estimate": 0.35204244179513766
+      "required_token_positions_per_second": 28400.437663185738,
+      "required_pages_per_second_upper_estimate": 0.3526383856263052
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1045449697734071,
-    "producer_p95_seconds": 0.17584805000024062,
-    "gpu_update_seconds": 0.9751540931992946,
-    "producer_batches_per_second": 9.565261744944978,
-    "gpu_batches_per_second": 1.0254789545303458,
-    "producer_headroom": 9.327604143105722,
+    "producer_batch_seconds": 0.053367566671880695,
+    "producer_p95_seconds": 0.17484046899971872,
+    "gpu_update_seconds": 0.9735061243735307,
+    "producer_batches_per_second": 18.737972562029867,
+    "gpu_batches_per_second": 1.0272149039057341,
+    "producer_headroom": 18.241531047479253,
     "sustainable": true,
-    "buffer_seconds": 60.459553778356266,
+    "buffer_seconds": 60.3573797111589,
     "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": 449255.4257474003,
+    "generation_bytes_per_second": 449162.1108758026,
     "upload_bytes_per_second": 6093636.035551512,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.39975623432643,
+    "estimated_full_shards_per_hour": 96.37973303514059,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 44700, "loss": 0.6488265637308359, "lr": 0.0003, "grad_norm": 0.13343255495650386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6541086829, "tokens_per_second": 147309.0517887333, "data_wait_s": 8.866999451129232e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 44800, "loss": 0.6016717441380024, "lr": 0.0003, "grad_norm": 0.1310367573175083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6555559996, "tokens_per_second": 147289.71960988585, "data_wait_s": 1.4893999832565896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 44900, "loss": 0.6635061390697956, "lr": 0.0003, "grad_norm": 0.1308439027483963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6570181205, "tokens_per_second": 149633.56542232283, "data_wait_s": 8.383000022149645e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 45000, "loss": 0.624427679926157, "lr": 0.0003, "grad_norm": 0.12727347341567702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6584864111, "tokens_per_second": 149770.12722589058, "data_wait_s": 7.5930001912638545e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13920758c8c2f1aa52a914fa0d71bd63600f5e4e712e50211ff7ec9378060365
-size 103947
+oid sha256:c01171e415e31dd0cfd68906c3f40027159865b899614d36220b45cda46dad0e
+size 106065