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": 75348.10300265734,
-      "required_pages_per_second_upper_estimate": 0.006530982600249193
+      "required_token_positions_per_second": 75481.9964301487,
+      "required_pages_per_second_upper_estimate": 0.006542588143194354
     },
     {
       "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": 9418.512875332168,
-      "required_pages_per_second_upper_estimate": 0.0037571856052864877
+      "required_token_positions_per_second": 9435.249553768588,
+      "required_pages_per_second_upper_estimate": 0.003763862116550418
     },
     {
       "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": 18837.025750664336,
-      "required_pages_per_second_upper_estimate": 0.004357614387085808
+      "required_token_positions_per_second": 18870.499107537176,
+      "required_pages_per_second_upper_estimate": 0.004365357859087379
     },
     {
       "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": 28255.538625996505,
-      "required_pages_per_second_upper_estimate": 0.0017206140569690212
+      "required_token_positions_per_second": 28305.748661305763,
+      "required_pages_per_second_upper_estimate": 0.0017236715846876587
     },
     {
       "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": 28255.538625996505,
-      "required_pages_per_second_upper_estimate": 0.002175189579766196
+      "required_token_positions_per_second": 28305.748661305763,
+      "required_pages_per_second_upper_estimate": 0.002179054887274576
     },
     {
       "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": 28255.538625996505,
-      "required_pages_per_second_upper_estimate": 0.3508392245302967
+      "required_token_positions_per_second": 28305.748661305763,
+      "required_pages_per_second_upper_estimate": 0.35146266512665936
     }
   ],
   "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": 0.4805245916249419,
-    "producer_p95_seconds": 0.2879444809987035,
-    "gpu_update_seconds": 0.9784984234759008,
-    "producer_batches_per_second": 2.081058945637725,
-    "gpu_batches_per_second": 1.0219740533129522,
-    "producer_headroom": 2.0363128974669342,
+    "producer_batch_seconds": 0.1836974928437769,
+    "producer_p95_seconds": 0.34346510599789326,
+    "gpu_update_seconds": 0.9767627180903746,
+    "producer_batches_per_second": 5.443732434881062,
+    "gpu_batches_per_second": 1.0237900991502373,
+    "producer_headroom": 5.317234889651159,
     "sustainable": true,
-    "buffer_seconds": 60.666902255505846,
+    "buffer_seconds": 60.55928852160323,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395879.0932043917,
-    "upload_bytes_per_second": 5677586.250118114,
+    "generation_bytes_per_second": 391678.9660212969,
+    "upload_bytes_per_second": 5730388.830556262,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.9464378080255,
+    "estimated_full_shards_per_hour": 84.04518828849011,
     "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": 1.3182481060011924,
-      "bytes": 8487979,
-      "pages": 27,
-      "shards": 4
-    },
     {
       "seconds": 1.0571894279983098,
       "bytes": 6049663,
@@ -562,6 +556,12 @@
       "bytes": 7683890,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1247381559987844,
+      "bytes": 11305256,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 22300, "loss": 0.7413192056119442, "lr": 0.0003, "grad_norm": 0.15615766636443054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3262803305, "tokens_per_second": 149942.17956689597, "data_wait_s": 1.29449981614016e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22400, "loss": 0.6735562570393085, "lr": 0.0003, "grad_norm": 0.13137382067411008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3277561472, "tokens_per_second": 149637.9703021123, "data_wait_s": 1.3070002751192078e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22500, "loss": 0.6838201712816954, "lr": 0.0003, "grad_norm": 0.1324805219547243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3292082005, "tokens_per_second": 132693.18922523074, "data_wait_s": 8.938997780205682e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 22600, "loss": 0.6810497716069221, "lr": 0.0003, "grad_norm": 0.13040007951598498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3306563191, "tokens_per_second": 146442.26679563854, "data_wait_s": 1.208100002259016e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7b7786a937f0563979d62e43bb9139c37b978eba68667ec86a14689770dd7e96
-size 474073
+oid sha256:7ad3311fe04289d4edef96fb0c91ae9175a7c16841a6d7885e0e30869c14587f
+size 476191