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": 75250.87497567486,
-      "required_pages_per_second_upper_estimate": 0.006522555121292533
+      "required_token_positions_per_second": 76111.26401445507,
+      "required_pages_per_second_upper_estimate": 0.0065971314625379186
     },
     {
       "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": 9406.359371959357,
-      "required_pages_per_second_upper_estimate": 0.0037523373910800053
+      "required_token_positions_per_second": 9513.908001806883,
+      "required_pages_per_second_upper_estimate": 0.003795240147521495
     },
     {
       "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": 18812.718743918715,
-      "required_pages_per_second_upper_estimate": 0.004351991388863916
+      "required_token_positions_per_second": 19027.816003613767,
+      "required_pages_per_second_upper_estimate": 0.004401750354311889
     },
     {
       "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": 28219.078115878074,
-      "required_pages_per_second_upper_estimate": 0.0017183938031963214
+      "required_token_positions_per_second": 28541.724005420652,
+      "required_pages_per_second_upper_estimate": 0.001738041245077308
     },
     {
       "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": 28219.078115878074,
-      "required_pages_per_second_upper_estimate": 0.0021723827487681274
+      "required_token_positions_per_second": 28541.724005420652,
+      "required_pages_per_second_upper_estimate": 0.002197220922486101
     },
     {
       "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": 28219.078115878074,
-      "required_pages_per_second_upper_estimate": 0.3503865070201035
+      "required_token_positions_per_second": 28541.724005420652,
+      "required_pages_per_second_upper_estimate": 0.3543926891418932
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07090172336737055,
-    "producer_p95_seconds": 0.27136119900023914,
-    "gpu_update_seconds": 0.9797626941059869,
-    "producer_batches_per_second": 14.10402952857147,
-    "gpu_batches_per_second": 1.0206553137976735,
-    "producer_headroom": 13.818601968663575,
+    "producer_batch_seconds": 0.18855529369528767,
+    "producer_p95_seconds": 0.2935596390016144,
+    "gpu_update_seconds": 0.9686871050518562,
+    "producer_batches_per_second": 5.303484088948661,
+    "gpu_batches_per_second": 1.032325087001615,
+    "producer_headroom": 5.13741664881226,
     "sustainable": true,
-    "buffer_seconds": 60.74528703457119,
+    "buffer_seconds": 60.05860051321508,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540875.5110972658,
-    "upload_bytes_per_second": 5047284.015228411,
+    "generation_bytes_per_second": 663227.07556234,
+    "upload_bytes_per_second": 5113595.432457817,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.05929374397735,
+    "estimated_full_shards_per_hour": 71.15654563976598,
     "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": 0.8513057910004136,
-      "bytes": 341762,
-      "pages": 9,
-      "shards": 1
-    },
     {
       "seconds": 1.0242165819981892,
       "bytes": 5354140,
@@ -562,6 +556,12 @@
       "bytes": 6141256,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8550289739978325,
+      "bytes": 5835711,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -227,3 +227,4 @@
 {"step": 62700, "loss": 0.6200138367712498, "lr": 0.0003, "grad_norm": 0.1287356247497853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9176141987, "tokens_per_second": 150068.7338120172, "data_wait_s": 1.2479002180043608e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 62800, "loss": 0.6389019126072526, "lr": 0.0003, "grad_norm": 0.1355427224723376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9190803656, "tokens_per_second": 148913.4956377439, "data_wait_s": 1.2069998774677515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 62900, "loss": 0.6282933112233877, "lr": 0.0003, "grad_norm": 0.12910738432293123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9205508537, "tokens_per_second": 150436.70115853802, "data_wait_s": 8.184000762412325e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 63000, "loss": 0.6200405359268188, "lr": 0.0003, "grad_norm": 0.12977827007988665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9220130946, "tokens_per_second": 148037.63914714358, "data_wait_s": 1.4934001228539273e-05, "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:f5e86b5198da91bbd3015bad61a68f391eb7d19254ec313c5115aa346189fa8b
-size 488319
+oid sha256:df3404e047e03da4892a5f4665e5238cda2561935612c6d2440bcf495953ad60
+size 490437