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": 75540.14388598053,
-      "required_pages_per_second_upper_estimate": 0.006547628217292478
+      "required_token_positions_per_second": 74714.86791579654,
+      "required_pages_per_second_upper_estimate": 0.006476095387839753
     },
     {
       "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": 9442.517985747567,
-      "required_pages_per_second_upper_estimate": 0.0037667616027395752
+      "required_token_positions_per_second": 9339.358489474567,
+      "required_pages_per_second_upper_estimate": 0.003725609737304359
     },
     {
       "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": 18885.035971495134,
-      "required_pages_per_second_upper_estimate": 0.0043687207067239844
+      "required_token_positions_per_second": 18678.716978949134,
+      "required_pages_per_second_upper_estimate": 0.00432099243888872
     },
     {
       "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": 28327.553957242704,
-      "required_pages_per_second_upper_estimate": 0.0017249994128066724
+      "required_token_positions_per_second": 28018.075468423704,
+      "required_pages_per_second_upper_estimate": 0.0017061537965457393
     },
     {
       "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": 28327.553957242704,
-      "required_pages_per_second_upper_estimate": 0.0021807335193167244
+      "required_token_positions_per_second": 28018.075468423704,
+      "required_pages_per_second_upper_estimate": 0.00215690900855615
     },
     {
       "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": 28327.553957242704,
-      "required_pages_per_second_upper_estimate": 0.35173341392456514
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28018.075468423704,
+      "required_pages_per_second_upper_estimate": 0.3478907268513069
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07929377635149137,
-    "producer_p95_seconds": 0.22303854299934756,
-    "gpu_update_seconds": 0.9760108494270838,
-    "producer_batches_per_second": 12.611330245733615,
-    "gpu_batches_per_second": 1.0245787744951789,
-    "producer_headroom": 12.30879514554394,
+    "producer_batch_seconds": 0.4606873029764387,
+    "producer_p95_seconds": 0.2853082700003142,
+    "gpu_update_seconds": 0.9867915457347962,
+    "producer_batches_per_second": 2.170669765672148,
+    "gpu_batches_per_second": 1.0133852527641676,
+    "producer_headroom": 2.141998573347407,
     "sustainable": true,
-    "buffer_seconds": 60.5126726644792,
+    "buffer_seconds": 60.19428428982257,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 497669.87662039424,
-    "upload_bytes_per_second": 4017531.564012735,
+    "generation_bytes_per_second": 422487.2191202642,
+    "upload_bytes_per_second": 4007153.86142561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.78837036093589,
+    "estimated_full_shards_per_hour": 90.6559222241015,
     "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": 0.9117075719987042,
-      "bytes": 2359249,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9851995050012192,
       "bytes": 5930447,
@@ -562,6 +556,12 @@
       "bytes": 5919931,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.7547573850006302,
+      "bytes": 662301,
+      "pages": 19,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -151,3 +151,4 @@
 {"step": 330100, "loss": 0.6080137956887484, "lr": 0.0003, "grad_norm": 0.17845473445853496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48317901815, "tokens_per_second": 134522.147590586, "data_wait_s": 1.3312999726622365e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330200, "loss": 0.6012338791042566, "lr": 0.0003, "grad_norm": 0.15536010491393057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48332406906, "tokens_per_second": 147673.95477579493, "data_wait_s": 1.365899879601784e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330300, "loss": 0.6037442348897457, "lr": 0.0003, "grad_norm": 0.17942496505797634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48346976699, "tokens_per_second": 140060.01440860407, "data_wait_s": 1.7827998817665502e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 330400, "loss": 0.6018347283825278, "lr": 0.0003, "grad_norm": 0.1707738632691945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48361619115, "tokens_per_second": 148133.64923696552, "data_wait_s": 1.0896999810938723e-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 sha256:2298564b758535726802ee86aec44a97054efde8598c42f925f0193e9542f6a9
-size 326568
+oid sha256:6c1966257f3663f4dd34ad307920844534e831afa737dd7f29fe7625b1e3b02d
+size 328686