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": 75159.57712292268,
-      "required_pages_per_second_upper_estimate": 0.0065146416548082665
+      "required_token_positions_per_second": 73723.53401031265,
+      "required_pages_per_second_upper_estimate": 0.006390169077425223
     },
     {
       "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": 9394.947140365335,
-      "required_pages_per_second_upper_estimate": 0.003747784881269082
+      "required_token_positions_per_second": 9215.441751289081,
+      "required_pages_per_second_upper_estimate": 0.003676177497721829
     },
     {
       "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": 18789.89428073067,
-      "required_pages_per_second_upper_estimate": 0.004346711351002195
+      "required_token_positions_per_second": 18430.883502578163,
+      "required_pages_per_second_upper_estimate": 0.004263660525850508
     },
     {
       "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": 28184.841421096007,
-      "required_pages_per_second_upper_estimate": 0.0017163089681101488
+      "required_token_positions_per_second": 27646.325253867246,
+      "required_pages_per_second_upper_estimate": 0.00168351615890189
     },
     {
       "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": 28184.841421096007,
-      "required_pages_per_second_upper_estimate": 0.002169747113230567
+      "required_token_positions_per_second": 27646.325253867246,
+      "required_pages_per_second_upper_estimate": 0.0021282906479691364
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28184.841421096007,
-      "required_pages_per_second_upper_estimate": 0.34996140185375674
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27646.325253867246,
+      "required_pages_per_second_upper_estimate": 0.34327483335444886
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0892034768907024,
-    "producer_p95_seconds": 0.3635556219960563,
-    "gpu_update_seconds": 0.9809528315921555,
-    "producer_batches_per_second": 11.210325369102614,
-    "gpu_batches_per_second": 1.0194170074181137,
-    "producer_headroom": 10.996800413890586,
+    "producer_batch_seconds": 0.16508619677313163,
+    "producer_p95_seconds": 0.34143745400069747,
+    "gpu_update_seconds": 1.0000605775312768,
+    "producer_batches_per_second": 6.057441624718279,
+    "gpu_batches_per_second": 0.9999394261381382,
+    "producer_headroom": 6.057808569577758,
     "sustainable": true,
-    "buffer_seconds": 60.81907555871364,
+    "buffer_seconds": 60.0036346518766,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 235077.345410548,
-    "upload_bytes_per_second": 5063403.408782123,
+    "generation_bytes_per_second": 258849.66542778863,
+    "upload_bytes_per_second": 5171341.662781372,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.442126004575066,
+    "estimated_full_shards_per_hour": 55.54311248898739,
     "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": 3.0481821479988866,
-      "bytes": 2407311,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.2172690299994429,
       "bytes": 6177200,
@@ -562,6 +556,12 @@
       "bytes": 10409429,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1135386019959697,
+      "bytes": 1273006,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 270300, "loss": 0.6106796804815531, "lr": 0.0003, "grad_norm": 0.14322447692362245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39563684915, "tokens_per_second": 120176.25725778555, "data_wait_s": 1.2378004612401128e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270400, "loss": 0.5690526589751244, "lr": 0.0003, "grad_norm": 0.14620697842648134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39578450717, "tokens_per_second": 149824.5612563105, "data_wait_s": 1.4336997992359102e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270500, "loss": 0.5988229159265757, "lr": 0.0003, "grad_norm": 0.1474321069976209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39593205699, "tokens_per_second": 150691.4202247918, "data_wait_s": 1.7910002497956157e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 270600, "loss": 0.58234653621912, "lr": 0.0003, "grad_norm": 0.15620423272467027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39607869036, "tokens_per_second": 149560.0738369881, "data_wait_s": 1.1085001460742205e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4357975de6d8fa6282efc0142a8814d513b31772fa7b3f662eeabedea8253fc7
-size 713058
+oid sha256:ec5f6cc23f264fb6d7e848541d7564f6bb7cbb7c4649f1727d9f0b051c01d343
+size 715176