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": 76205.01309604496,
-      "required_pages_per_second_upper_estimate": 0.006605257395325256
+      "required_token_positions_per_second": 75483.18640121476,
+      "required_pages_per_second_upper_estimate": 0.006542691286870406
     },
     {
       "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": 9525.62663700562,
-      "required_pages_per_second_upper_estimate": 0.003799914886311481
+      "required_token_positions_per_second": 9435.398300151845,
+      "required_pages_per_second_upper_estimate": 0.0037639214537066555
     },
     {
       "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": 19051.25327401124,
-      "required_pages_per_second_upper_estimate": 0.004407172154336474
+      "required_token_positions_per_second": 18870.79660030369,
+      "required_pages_per_second_upper_estimate": 0.004365426678803218
     },
     {
       "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": 28576.879911016862,
-      "required_pages_per_second_upper_estimate": 0.0017401820552793354
+      "required_token_positions_per_second": 28306.194900455535,
+      "required_pages_per_second_upper_estimate": 0.0017236987583106442
     },
     {
       "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": 28576.879911016862,
-      "required_pages_per_second_upper_estimate": 0.0021999273214166723
+      "required_token_positions_per_second": 28306.194900455535,
+      "required_pages_per_second_upper_estimate": 0.002179089239999595
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28576.879911016862,
-      "required_pages_per_second_upper_estimate": 0.3548292078301509
+      "required_token_positions_per_second": 28306.194900455535,
+      "required_pages_per_second_upper_estimate": 0.35146820592343314
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0798406615626277,
-    "producer_p95_seconds": 0.2157158840000193,
-    "gpu_update_seconds": 0.9674954048899242,
-    "producer_batches_per_second": 12.52494631717939,
-    "gpu_batches_per_second": 1.033596640300089,
-    "producer_headroom": 12.117828008364041,
+    "producer_batch_seconds": 0.07478907846098082,
+    "producer_p95_seconds": 0.20420235600067826,
+    "gpu_update_seconds": 0.9767473197026231,
+    "producer_batches_per_second": 13.370936246015159,
+    "gpu_batches_per_second": 1.0238062391657818,
+    "producer_headroom": 13.060026140209958,
     "sustainable": true,
-    "buffer_seconds": 60.95221050806523,
+    "buffer_seconds": 60.55833382156263,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393231.41264291684,
-    "upload_bytes_per_second": 6190803.877461106,
+    "generation_bytes_per_second": 345152.5905722208,
+    "upload_bytes_per_second": 6062939.340143687,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.37830719438199,
+    "estimated_full_shards_per_hour": 74.06171119570702,
     "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": 1.3945427510002446,
-      "bytes": 16977562,
-      "pages": 43,
-      "shards": 5
-    },
     {
       "seconds": 0.7539118750000853,
       "bytes": 789098,
@@ -562,6 +556,12 @@
       "bytes": 6106536,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1473440179997851,
+      "bytes": 6606039,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 50500, "loss": 0.5936776921153069, "lr": 0.0003, "grad_norm": 0.12213298037925359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7390598106, "tokens_per_second": 150642.64040251233, "data_wait_s": 1.0911999197560363e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 50600, "loss": 0.6595828458666801, "lr": 0.0003, "grad_norm": 0.14864370134429494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7405328376, "tokens_per_second": 151175.29936112068, "data_wait_s": 8.854000043356791e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 50700, "loss": 0.6321609169244766, "lr": 0.0003, "grad_norm": 0.12893147333020086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7419983847, "tokens_per_second": 148425.897060511, "data_wait_s": 9.569001122144982e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 50800, "loss": 0.6277779079973698, "lr": 0.0003, "grad_norm": 0.13406036670035498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7434572545, "tokens_per_second": 149314.30217967523, "data_wait_s": 8.113000149023719e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:637bc8bd023eb869d317426779b700572dbea6857c0c75e6e8afc46e26ad7bcf
-size 228357
+oid sha256:01e4016031c5f0d267fcb0a568909dfedb3af5f12a9dda59fc1daf1a813f0cf3
+size 230475