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": 73843.86885215543,
-      "required_pages_per_second_upper_estimate": 0.006400599396530286
+      "required_token_positions_per_second": 74400.6790362521,
+      "required_pages_per_second_upper_estimate": 0.006448862292065276
     },
     {
       "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": 9230.483606519429,
-      "required_pages_per_second_upper_estimate": 0.0036821779186689916
+      "required_token_positions_per_second": 9300.084879531512,
+      "required_pages_per_second_upper_estimate": 0.0037099429071052784
     },
     {
       "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": 18460.967213038857,
-      "required_pages_per_second_upper_estimate": 0.004270619862810369
+      "required_token_positions_per_second": 18600.169759063025,
+      "required_pages_per_second_upper_estimate": 0.004302821921951924
     },
     {
       "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": 27691.450819558286,
-      "required_pages_per_second_upper_estimate": 0.0016862640690969287
+      "required_token_positions_per_second": 27900.25463859454,
+      "required_pages_per_second_upper_estimate": 0.0016989791261672635
     },
     {
       "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": 27691.450819558286,
-      "required_pages_per_second_upper_estimate": 0.0021317645389315938
+      "required_token_positions_per_second": 27900.25463859454,
+      "required_pages_per_second_upper_estimate": 0.0021478388349269696
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27691.450819558286,
-      "required_pages_per_second_upper_estimate": 0.34383514185477837
+      "required_token_positions_per_second": 27900.25463859454,
+      "required_pages_per_second_upper_estimate": 0.3464277864657802
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10607608259384449,
-    "producer_p95_seconds": 0.2160918280005717,
-    "gpu_update_seconds": 0.9984308940748025,
-    "producer_batches_per_second": 9.42719579708564,
-    "gpu_batches_per_second": 1.0015715718879588,
-    "producer_headroom": 9.412403528302436,
+    "producer_batch_seconds": 0.09630696202353306,
+    "producer_p95_seconds": 0.22131155799979751,
+    "gpu_update_seconds": 0.9909586976225804,
+    "producer_batches_per_second": 10.383465317446575,
+    "gpu_batches_per_second": 1.0091237933519437,
+    "producer_headroom": 10.28958526778609,
     "sustainable": true,
-    "buffer_seconds": 60.90428453856295,
+    "buffer_seconds": 60.4484805549774,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391320.6513255723,
-    "upload_bytes_per_second": 4967026.690318246,
+    "generation_bytes_per_second": 375435.26413780276,
+    "upload_bytes_per_second": 4960880.305291673,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.96830229592682,
+    "estimated_full_shards_per_hour": 80.55966799831927,
     "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": 2.0074904029997924,
-      "bytes": 13527574,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 1.034237903999383,
       "bytes": 890095,
@@ -562,6 +556,12 @@
       "bytes": 5902213,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.6428124459998799,
+      "bytes": 11197872,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -122,3 +122,4 @@
 {"step": 249300, "loss": 0.5752866845577955, "lr": 0.0003, "grad_norm": 0.16293882183368283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36491026020, "tokens_per_second": 150820.65796480002, "data_wait_s": 1.0529000064707361e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249400, "loss": 0.6120243947952986, "lr": 0.0003, "grad_norm": 0.14405296487039118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36505670984, "tokens_per_second": 149597.77159190035, "data_wait_s": 1.2959999367012642e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249500, "loss": 0.5780062265694141, "lr": 0.0003, "grad_norm": 0.1606750468806231, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36520295531, "tokens_per_second": 147778.30449327236, "data_wait_s": 8.953998985816725e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 249600, "loss": 0.6521867774426937, "lr": 0.0003, "grad_norm": 0.20672822852499556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36534749015, "tokens_per_second": 147591.0138329264, "data_wait_s": 1.2995000361115672e-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 sha256:2798aea89408e66be83ed811f372890da0283547a0f0defc1032ce1ec0ec8db5
-size 264363
+oid sha256:5ddc9ccda031263b79dfae3d743ed4003b929d8d48312feb44bb96cfb2bcd2ec
+size 266481