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": 77069.6996497935,
-      "required_pages_per_second_upper_estimate": 0.005331860648113364
+      "required_token_positions_per_second": 76216.25280854134,
+      "required_pages_per_second_upper_estimate": 0.005272817215366039
     },
     {
       "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": 9633.712456224188,
-      "required_pages_per_second_upper_estimate": 0.004033434230628633
+      "required_token_positions_per_second": 9527.031601067667,
+      "required_pages_per_second_upper_estimate": 0.00398876918432418
     },
     {
       "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": 19267.424912448376,
-      "required_pages_per_second_upper_estimate": 0.0040128964930987235
+      "required_token_positions_per_second": 19054.063202135334,
+      "required_pages_per_second_upper_estimate": 0.003968458875567199
     },
     {
       "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": 28901.137368672564,
-      "required_pages_per_second_upper_estimate": 0.0017325790557703382
+      "required_token_positions_per_second": 28581.094803203003,
+      "required_pages_per_second_upper_estimate": 0.00171339299264714
     },
     {
       "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": 28901.137368672564,
-      "required_pages_per_second_upper_estimate": 0.002246936307745745
+      "required_token_positions_per_second": 28581.094803203003,
+      "required_pages_per_second_upper_estimate": 0.0022220544059989595
     },
     {
       "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": 28901.137368672564,
-      "required_pages_per_second_upper_estimate": 0.37071275854173963
+      "required_token_positions_per_second": 28581.094803203003,
+      "required_pages_per_second_upper_estimate": 0.36660759614682986
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11592747762534827,
-    "producer_p95_seconds": 0.35787701699882746,
-    "gpu_update_seconds": 0.9566405517994976,
-    "producer_batches_per_second": 8.626082620651651,
-    "gpu_batches_per_second": 1.0453247022812704,
-    "producer_headroom": 8.252060438088252,
+    "producer_batch_seconds": 0.12341628803909543,
+    "producer_p95_seconds": 0.350743163995503,
+    "gpu_update_seconds": 0.9673527270516442,
+    "producer_batches_per_second": 8.102658213826874,
+    "gpu_batches_per_second": 1.033749088657516,
+    "producer_headroom": 7.838128519512831,
     "sustainable": true,
-    "buffer_seconds": 60.26835476336835,
+    "buffer_seconds": 60.94322180425358,
     "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": 353544.9060392489,
-    "upload_bytes_per_second": 4639888.393057113,
+    "generation_bytes_per_second": 402220.5895584841,
+    "upload_bytes_per_second": 4575897.294281291,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.86250673182583,
+    "estimated_full_shards_per_hour": 86.30717530313389,
     "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": 1.400596306000807,
-      "bytes": 7658131,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.01357181700223,
       "bytes": 644212,
@@ -562,6 +556,12 @@
       "bytes": 6041226,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.2637325059986324,
+      "bytes": 992368,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -405,3 +405,4 @@
 {"step": 644900, "loss": 0.48915520310401917, "lr": 0.0001, "grad_norm": 0.18857827802658592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94396571499, "tokens_per_second": 151698.3424844779, "data_wait_s": 1.4887998986523598e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645000, "loss": 0.5380465984344482, "lr": 0.0001, "grad_norm": 0.1812405295203612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94411289736, "tokens_per_second": 151737.97812397123, "data_wait_s": 1.7938000382855535e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645100, "loss": 0.5672286152839661, "lr": 0.0001, "grad_norm": 0.19267931109074643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94425964227, "tokens_per_second": 143045.67401698208, "data_wait_s": 1.160599640570581e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 645200, "loss": 0.5682960152626038, "lr": 0.0001, "grad_norm": 0.18092363405899495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94440612732, "tokens_per_second": 152416.5722103898, "data_wait_s": 1.475700264563784e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:82d1f486f210ff7408761ae0fbf9bc6201e3e624809bddb32b42795172b1fa00
-size 869238
+oid sha256:3b13839384fc2d991fe7967ac1c4a36a1ce324da2f7a4008a7b331defd6338bd
+size 871356