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": 74887.82545313798,
-      "required_pages_per_second_upper_estimate": 0.0064910869088196385
+      "required_token_positions_per_second": 74070.13649864245,
+      "required_pages_per_second_upper_estimate": 0.006420211702657674
     },
     {
       "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": 9360.978181642247,
-      "required_pages_per_second_upper_estimate": 0.00373423415575325
+      "required_token_positions_per_second": 9258.767062330306,
+      "required_pages_per_second_upper_estimate": 0.003693460612067299
     },
     {
       "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": 18721.956363284495,
-      "required_pages_per_second_upper_estimate": 0.004330995109467532
+      "required_token_positions_per_second": 18517.53412466061,
+      "required_pages_per_second_upper_estimate": 0.004283705622270419
     },
     {
       "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": 28082.934544926742,
-      "required_pages_per_second_upper_estimate": 0.0017101033740154988
+      "required_token_positions_per_second": 27776.301186990917,
+      "required_pages_per_second_upper_estimate": 0.001691431011298103
     },
     {
       "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": 28082.934544926742,
-      "required_pages_per_second_upper_estimate": 0.002161902039806769
+      "required_token_positions_per_second": 27776.301186990917,
+      "required_pages_per_second_upper_estimate": 0.0021382965550974077
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28082.934544926742,
-      "required_pages_per_second_upper_estimate": 0.34869605951210925
+      "required_token_positions_per_second": 27776.301186990917,
+      "required_pages_per_second_upper_estimate": 0.3448886994423795
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09758196250007245,
-    "producer_p95_seconds": 0.34619067899984657,
-    "gpu_update_seconds": 0.9845124965757784,
-    "producer_batches_per_second": 10.247795539050134,
-    "gpu_batches_per_second": 1.015731139501112,
-    "producer_headroom": 10.089082770548373,
+    "producer_batch_seconds": 0.10509689617197182,
+    "producer_p95_seconds": 0.3456709589991078,
+    "gpu_update_seconds": 0.9953809117302124,
+    "producer_batches_per_second": 9.515028858356418,
+    "gpu_batches_per_second": 1.0046405232563267,
+    "producer_headroom": 9.471078100170093,
     "sustainable": true,
-    "buffer_seconds": 60.05526229112248,
+    "buffer_seconds": 60.71823561554295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283006.05425264407,
-    "upload_bytes_per_second": 3969276.193098734,
+    "generation_bytes_per_second": 312462.092878963,
+    "upload_bytes_per_second": 3931766.5556723503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.7265111988496,
+    "estimated_full_shards_per_hour": 67.04709138657252,
     "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": 1.0817985579997185,
-      "bytes": 6075598,
-      "pages": 20,
-      "shards": 2
-    },
     {
       "seconds": 1.3699396060001163,
       "bytes": 12184416,
@@ -562,6 +556,12 @@
       "bytes": 7655478,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.7887198950011225,
+      "bytes": 905384,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 140800, "loss": 0.5893173590302467, "lr": 0.0003, "grad_norm": 0.13309273764161786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20606050889, "tokens_per_second": 149140.1489081394, "data_wait_s": 1.3107997801853344e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140900, "loss": 0.6342130172997713, "lr": 0.0003, "grad_norm": 0.15846270861170156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20620749311, "tokens_per_second": 149953.54148665606, "data_wait_s": 1.0063002264359966e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141000, "loss": 0.6059113256633282, "lr": 0.0003, "grad_norm": 0.17628415937895242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20635479752, "tokens_per_second": 149509.69839055766, "data_wait_s": 1.4297002053353935e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 141100, "loss": 0.590720796957612, "lr": 0.0003, "grad_norm": 0.1466954825012965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20649955142, "tokens_per_second": 148243.05961548552, "data_wait_s": 1.7220998415723443e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:18c735e520ba04cd785b841119798317f4a80b527c5687bc48d11119c33000af
-size 554760
+oid sha256:8cdf4b7a5054749270165b83979df4cc43894c9ddb9a2264252c6b78628df984
+size 556878