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": 74470.85442313964,
-      "required_pages_per_second_upper_estimate": 0.006454944916742796
+      "required_token_positions_per_second": 75703.9831219617,
+      "required_pages_per_second_upper_estimate": 0.006561829387020579
     },
     {
       "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": 9308.856802892455,
-      "required_pages_per_second_upper_estimate": 0.0037134421584859
+      "required_token_positions_per_second": 9462.997890245213,
+      "required_pages_per_second_upper_estimate": 0.0037749313428455454
     },
     {
       "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": 18617.71360578491,
-      "required_pages_per_second_upper_estimate": 0.0043068803821298745
+      "required_token_positions_per_second": 18925.995780490426,
+      "required_pages_per_second_upper_estimate": 0.004378196037667028
     },
     {
       "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": 27926.570408677366,
-      "required_pages_per_second_upper_estimate": 0.0017005816184971325
+      "required_token_positions_per_second": 28388.99367073564,
+      "required_pages_per_second_upper_estimate": 0.0017287407690091277
     },
     {
       "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": 27926.570408677366,
-      "required_pages_per_second_upper_estimate": 0.0021498646957546596
+      "required_token_positions_per_second": 28388.99367073564,
+      "required_pages_per_second_upper_estimate": 0.002185463318537423
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27926.570408677366,
-      "required_pages_per_second_upper_estimate": 0.34675454025699204
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28388.99367073564,
+      "required_pages_per_second_upper_estimate": 0.3524962895406539
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08801844307808437,
-    "producer_p95_seconds": 0.2268695309976465,
-    "gpu_update_seconds": 0.9900248972716389,
-    "producer_batches_per_second": 11.361255266840649,
-    "gpu_batches_per_second": 1.0100756079527404,
-    "producer_headroom": 11.247925578430781,
+    "producer_batch_seconds": 0.22740758702343555,
+    "producer_p95_seconds": 0.31379283199930796,
+    "gpu_update_seconds": 0.973898558035205,
+    "producer_batches_per_second": 4.397390663561919,
+    "gpu_batches_per_second": 1.0268009863547323,
+    "producer_headroom": 4.2826124263604255,
     "sustainable": true,
-    "buffer_seconds": 60.391518733569974,
+    "buffer_seconds": 60.38171059818271,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478363.89569642034,
-    "upload_bytes_per_second": 5748522.218575571,
+    "generation_bytes_per_second": 446325.6246867401,
+    "upload_bytes_per_second": 5859113.273311393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.64575627488573,
+    "estimated_full_shards_per_hour": 95.77108912898686,
     "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": 0.8556265790011821,
-      "bytes": 1138175,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1632403840012557,
       "bytes": 11705717,
@@ -562,6 +556,12 @@
       "bytes": 1147220,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2733968520005874,
+      "bytes": 11422716,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 330800, "loss": 0.5701847933232784, "lr": 0.0003, "grad_norm": 0.15614280457714352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48420285230, "tokens_per_second": 147885.39125797685, "data_wait_s": 1.1126001481898129e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330900, "loss": 0.6047540437430143, "lr": 0.0003, "grad_norm": 0.142320208239958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48434964339, "tokens_per_second": 149440.5948079516, "data_wait_s": 1.091599915525876e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 331000, "loss": 0.5964784566313028, "lr": 0.0003, "grad_norm": 0.1577556859694088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48449730099, "tokens_per_second": 150677.61027192196, "data_wait_s": 1.105499904952012e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 331100, "loss": 0.6439957339316607, "lr": 0.0003, "grad_norm": 0.16588003596327064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48464504517, "tokens_per_second": 149608.35486879697, "data_wait_s": 1.7828999261837453e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:283ab7fcd5ca591e223a3bebb6b462dabfa19264d8cebd9b073d82eb1ffff87c
-size 341394
+oid sha256:bb92eff3728cce5afce0da2e2dfb24531225bf08241f1a18d21e2a0f7795f368
+size 343512