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": 77358.33009734105,
-      "required_pages_per_second_upper_estimate": 0.00535182877218961
+      "required_token_positions_per_second": 77027.31945272144,
+      "required_pages_per_second_upper_estimate": 0.005328928687744313
     },
     {
       "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": 9669.791262167631,
-      "required_pages_per_second_upper_estimate": 0.004048539673266012
+      "required_token_positions_per_second": 9628.41493159018,
+      "required_pages_per_second_upper_estimate": 0.004031216267689268
     },
     {
       "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": 19339.582524335263,
-      "required_pages_per_second_upper_estimate": 0.00402792502073057
+      "required_token_positions_per_second": 19256.82986318036,
+      "required_pages_per_second_upper_estimate": 0.004010689823746434
     },
     {
       "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": 29009.373786502896,
-      "required_pages_per_second_upper_estimate": 0.0017390676637518258
+      "required_token_positions_per_second": 28885.24479477054,
+      "required_pages_per_second_upper_estimate": 0.001731626320231467
     },
     {
       "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": 29009.373786502896,
-      "required_pages_per_second_upper_estimate": 0.002255351213150365
+      "required_token_positions_per_second": 28885.24479477054,
+      "required_pages_per_second_upper_estimate": 0.0022457007300289073
     },
     {
       "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": 29009.373786502896,
-      "required_pages_per_second_upper_estimate": 0.37210109909445616
+      "required_token_positions_per_second": 28885.24479477054,
+      "required_pages_per_second_upper_estimate": 0.37050890566784084
     }
   ],
   "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": 0.18922923376572953,
-    "producer_p95_seconds": 0.23168612999870675,
-    "gpu_update_seconds": 0.9530712452973977,
-    "producer_batches_per_second": 5.284595726038952,
-    "gpu_batches_per_second": 1.0492395032733974,
-    "producer_headroom": 5.036596229509249,
+    "producer_batch_seconds": 0.31906400767979903,
+    "producer_p95_seconds": 0.2318340800011356,
+    "gpu_update_seconds": 0.9571668925238074,
+    "producer_batches_per_second": 3.1341673643225954,
+    "gpu_batches_per_second": 1.044749884070115,
+    "producer_headroom": 2.99992123675819,
     "sustainable": true,
-    "buffer_seconds": 60.04348845373606,
+    "buffer_seconds": 60.301514228999864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 517633.62446598423,
-    "upload_bytes_per_second": 5155933.0149229895,
+    "generation_bytes_per_second": 467910.1287446776,
+    "upload_bytes_per_second": 5197529.626142483,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.07212591633458,
+    "estimated_full_shards_per_hour": 100.40262123828168,
     "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": 1.0131233649990463,
-      "bytes": 2204351,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.1051343430008274,
       "bytes": 10092152,
@@ -562,6 +556,12 @@
       "bytes": 7851564,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0146159530013392,
+      "bytes": 5636542,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 577700, "loss": 0.5548363924026489, "lr": 0.0003, "grad_norm": 0.16746688977543625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84558062867, "tokens_per_second": 153351.48063650445, "data_wait_s": 8.284001523861662e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 577800, "loss": 0.5663211941719055, "lr": 0.0003, "grad_norm": 0.19217156588652878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84572541026, "tokens_per_second": 150382.1330178825, "data_wait_s": 1.0037998436018825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 577900, "loss": 0.570175290107727, "lr": 0.0003, "grad_norm": 0.14820704823070996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84587141374, "tokens_per_second": 152481.7522259757, "data_wait_s": 1.3100001524435356e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 578000, "loss": 0.6239941716194153, "lr": 0.0003, "grad_norm": 0.1688119190484631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84601838982, "tokens_per_second": 152260.97375723397, "data_wait_s": 8.334998710779473e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f319834a595aee591d51a9e5ccc948bbebd21440b9c0225534e19d07ac34e932
-size 443841
+oid sha256:59eb49e6feae6d7c8f69ea540b5e0a556b498c03dd13df092c8c2af496cc10bc
+size 445959