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": 74400.6790362521,
-      "required_pages_per_second_upper_estimate": 0.006448862292065276
+      "required_token_positions_per_second": 75230.52059382049,
+      "required_pages_per_second_upper_estimate": 0.006520790855061105
     },
     {
       "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": 9300.084879531512,
-      "required_pages_per_second_upper_estimate": 0.0037099429071052784
+      "required_token_positions_per_second": 9403.815074227561,
+      "required_pages_per_second_upper_estimate": 0.003751322432674151
     },
     {
       "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": 18600.169759063025,
-      "required_pages_per_second_upper_estimate": 0.004302821921951924
+      "required_token_positions_per_second": 18807.630148455122,
+      "required_pages_per_second_upper_estimate": 0.004350814231859635
     },
     {
       "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": 27900.25463859454,
-      "required_pages_per_second_upper_estimate": 0.0016989791261672635
+      "required_token_positions_per_second": 28211.445222682683,
+      "required_pages_per_second_upper_estimate": 0.0017179290000474178
     },
     {
       "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": 27900.25463859454,
-      "required_pages_per_second_upper_estimate": 0.0021478388349269696
+      "required_token_positions_per_second": 28211.445222682683,
+      "required_pages_per_second_upper_estimate": 0.0021717951475207454
     },
     {
       "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": 27900.25463859454,
-      "required_pages_per_second_upper_estimate": 0.3464277864657802
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28211.445222682683,
+      "required_pages_per_second_upper_estimate": 0.35029173203226693
     }
   ],
   "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": 0.09630696202353306,
-    "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,
+    "producer_batch_seconds": 0.07286838666408357,
+    "producer_p95_seconds": 0.1938263989995903,
+    "gpu_update_seconds": 0.9800277788594234,
+    "producer_batches_per_second": 13.723372312466669,
+    "gpu_batches_per_second": 1.020379239825039,
+    "producer_headroom": 13.44928608584762,
     "sustainable": true,
-    "buffer_seconds": 60.4484805549774,
+    "buffer_seconds": 60.76172228928425,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375435.26413780276,
-    "upload_bytes_per_second": 4960880.305291673,
+    "generation_bytes_per_second": 392381.4397826354,
+    "upload_bytes_per_second": 4951636.511839129,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.55966799831927,
+    "estimated_full_shards_per_hour": 84.19592280492112,
     "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": 1.034237903999383,
-      "bytes": 890095,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.0843116149999332,
       "bytes": 11777166,
@@ -562,6 +556,12 @@
       "bytes": 11197872,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.1795239030016091,
+      "bytes": 829951,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -123,3 +123,4 @@
 {"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}
+{"step": 249700, "loss": 0.5862288670614362, "lr": 0.0003, "grad_norm": 0.14284561828266779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36549234595, "tokens_per_second": 148169.65117840783, "data_wait_s": 1.2750999303534627e-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 sha256:5ddc9ccda031263b79dfae3d743ed4003b929d8d48312feb44bb96cfb2bcd2ec
-size 266481
+oid sha256:6406f906b48099cc336055624c02fae2f9aa1c471aba59ec6b3460ac95fc0a8d
+size 268599