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": 77350.4927937165,
-      "required_pages_per_second_upper_estimate": 0.005351286569339814
+      "required_token_positions_per_second": 77269.82653885457,
+      "required_pages_per_second_upper_estimate": 0.005345705890656967
     },
     {
       "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": 9668.811599214563,
-      "required_pages_per_second_upper_estimate": 0.004048129508845251
+      "required_token_positions_per_second": 9658.728317356821,
+      "required_pages_per_second_upper_estimate": 0.004043907849294284
     },
     {
       "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": 19337.623198429126,
-      "required_pages_per_second_upper_estimate": 0.004027516944815221
+      "required_token_positions_per_second": 19317.456634713642,
+      "required_pages_per_second_upper_estimate": 0.004023316781421335
     },
     {
       "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": 29006.434797643687,
-      "required_pages_per_second_upper_estimate": 0.0017388914758572924
+      "required_token_positions_per_second": 28976.184952070464,
+      "required_pages_per_second_upper_estimate": 0.0017370780438040175
     },
     {
       "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": 29006.434797643687,
-      "required_pages_per_second_upper_estimate": 0.002255122719693807
+      "required_token_positions_per_second": 28976.184952070464,
+      "required_pages_per_second_upper_estimate": 0.002252770927255498
     },
     {
       "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": 29006.434797643687,
-      "required_pages_per_second_upper_estimate": 0.3720634009010106
+      "required_token_positions_per_second": 28976.184952070464,
+      "required_pages_per_second_upper_estimate": 0.3716753883617509
     }
   ],
   "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": 0.10375687899221475,
-    "producer_p95_seconds": 0.29647106600168627,
-    "gpu_update_seconds": 0.9531678123450719,
-    "producer_batches_per_second": 9.637915188977818,
-    "gpu_batches_per_second": 1.0491332030397746,
-    "producer_headroom": 9.186550536245326,
+    "producer_batch_seconds": 0.5453331487343291,
+    "producer_p95_seconds": 0.3108636420001858,
+    "gpu_update_seconds": 0.954162877056886,
+    "producer_batches_per_second": 1.8337414520296687,
+    "gpu_batches_per_second": 1.048039096935419,
+    "producer_headroom": 1.7496880196471003,
     "sustainable": true,
-    "buffer_seconds": 60.04957217773953,
+    "buffer_seconds": 60.112261254583814,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351894.5337970377,
-    "upload_bytes_per_second": 5451732.3201737655,
+    "generation_bytes_per_second": 298666.0467808024,
+    "upload_bytes_per_second": 5382940.222719963,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.50837526734684,
+    "estimated_full_shards_per_hour": 64.08678104942373,
     "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": 1.1795219960004033,
-      "bytes": 10882665,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.0230853959983506,
       "bytes": 6145347,
@@ -562,6 +556,12 @@
       "bytes": 6015320,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0212832630022604,
+      "bytes": 4591085,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -300,3 +300,4 @@
 {"step": 470200, "loss": 0.5777687430381775, "lr": 0.0003, "grad_norm": 0.16518298862429426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68823201498, "tokens_per_second": 153452.1597451511, "data_wait_s": 1.7800997738959268e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 470300, "loss": 0.593895435333252, "lr": 0.0003, "grad_norm": 0.19112237144961794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68837909629, "tokens_per_second": 152115.31681617472, "data_wait_s": 8.101000275928527e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 470400, "loss": 0.5647950768470764, "lr": 0.0003, "grad_norm": 0.18179823607544937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68852602581, "tokens_per_second": 153296.5250624911, "data_wait_s": 1.1593001545406878e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 470500, "loss": 0.5901936888694763, "lr": 0.0003, "grad_norm": 0.14091857068505795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68867103923, "tokens_per_second": 115552.92584162111, "data_wait_s": 1.7366000975016505e-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 sha256:570e77d7201028cab60ff46ae63ccc8feeeade8a34416be5943f2eafcec9a064
-size 644499
+oid sha256:f952133e58b5e30ec238b99158058a53109fc9fe0181e84233d2743fe6079313
+size 646617