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": 77466.2156031102,
-      "required_pages_per_second_upper_estimate": 0.005359292541807582
+      "required_token_positions_per_second": 76566.82580254077,
+      "required_pages_per_second_upper_estimate": 0.005297070668532336
     },
     {
       "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": 9683.276950388776,
-      "required_pages_per_second_upper_estimate": 0.004054185849311011
+      "required_token_positions_per_second": 9570.853225317596,
+      "required_pages_per_second_upper_estimate": 0.004007116383298051
     },
     {
       "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": 19366.55390077755,
-      "required_pages_per_second_upper_estimate": 0.00403354244716047
+      "required_token_positions_per_second": 19141.706450635193,
+      "required_pages_per_second_upper_estimate": 0.003986712652921828
     },
     {
       "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": 29049.830851166327,
-      "required_pages_per_second_upper_estimate": 0.0017414930030040375
+      "required_token_positions_per_second": 28712.559675952787,
+      "required_pages_per_second_upper_estimate": 0.0017212741110332519
     },
     {
       "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": 29049.830851166327,
-      "required_pages_per_second_upper_estimate": 0.0022584965719761258
+      "required_token_positions_per_second": 28712.559675952787,
+      "required_pages_per_second_upper_estimate": 0.002232275221602392
     },
     {
       "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": 29049.830851166327,
-      "required_pages_per_second_upper_estimate": 0.372620038880547
+      "required_token_positions_per_second": 28712.559675952787,
+      "required_pages_per_second_upper_estimate": 0.36829388637848137
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09470696961687963,
-    "producer_p95_seconds": 0.357535902003292,
-    "gpu_update_seconds": 0.9517439237994721,
-    "producer_batches_per_second": 10.558884990675173,
-    "gpu_batches_per_second": 1.0507027940960043,
-    "producer_headroom": 10.049354631972543,
+    "producer_batch_seconds": 0.1526549136489166,
+    "producer_p95_seconds": 0.34291676299471874,
+    "gpu_update_seconds": 0.9629235537351143,
+    "producer_batches_per_second": 6.550722646896582,
+    "gpu_batches_per_second": 1.0385040392054683,
+    "producer_headroom": 6.30784513068275,
     "sustainable": true,
-    "buffer_seconds": 60.91161112316622,
+    "buffer_seconds": 60.6641838853122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417865.84148640657,
-    "upload_bytes_per_second": 5060448.273915253,
+    "generation_bytes_per_second": 360116.71661703644,
+    "upload_bytes_per_second": 5065409.213739434,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.66428216404103,
+    "estimated_full_shards_per_hour": 77.27266429789849,
     "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": 1.0639040940004634,
-      "bytes": 11047745,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.0850444019997667,
       "bytes": 7634525,
@@ -562,6 +556,12 @@
       "bytes": 963026,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.1533947639982216,
+      "bytes": 11916585,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -388,3 +388,4 @@
 {"step": 479000, "loss": 0.5465214252471924, "lr": 0.0003, "grad_norm": 0.1953278472004877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70110154347, "tokens_per_second": 151701.75617601397, "data_wait_s": 1.2450997019186616e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 479100, "loss": 0.5731852650642395, "lr": 0.0003, "grad_norm": 0.16943549593553633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70124693880, "tokens_per_second": 150920.49653011974, "data_wait_s": 1.201400300487876e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 479200, "loss": 0.6261307597160339, "lr": 0.0003, "grad_norm": 0.1580683186868386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70139448421, "tokens_per_second": 148096.52265604865, "data_wait_s": 1.2021002476103604e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 479300, "loss": 0.5835630893707275, "lr": 0.0003, "grad_norm": 0.15958728987571263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70154115165, "tokens_per_second": 152735.8935683535, "data_wait_s": 1.2071999663021415e-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 sha256:216f0c80c5e39b4cc31868fb75231d4b86dd9246690c125859100c95b75270fc
-size 831666
+oid sha256:fa78fc099785c57773270386cb3fbbef2c8905b4d7ef9a2fd24e7f5a69d42a9c
+size 833784