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": 75230.52059382049,
-      "required_pages_per_second_upper_estimate": 0.006520790855061105
+      "required_token_positions_per_second": 75625.92984830218,
+      "required_pages_per_second_upper_estimate": 0.006555063927083985
     },
     {
       "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": 9403.815074227561,
-      "required_pages_per_second_upper_estimate": 0.003751322432674151
+      "required_token_positions_per_second": 9453.241231037773,
+      "required_pages_per_second_upper_estimate": 0.003771039265612643
     },
     {
       "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": 18807.630148455122,
-      "required_pages_per_second_upper_estimate": 0.004350814231859635
+      "required_token_positions_per_second": 18906.482462075546,
+      "required_pages_per_second_upper_estimate": 0.004373681974874421
     },
     {
       "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": 28211.445222682683,
-      "required_pages_per_second_upper_estimate": 0.0017179290000474178
+      "required_token_positions_per_second": 28359.723693113316,
+      "required_pages_per_second_upper_estimate": 0.0017269583809396325
     },
     {
       "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": 28211.445222682683,
-      "required_pages_per_second_upper_estimate": 0.0021717951475207454
+      "required_token_positions_per_second": 28359.723693113316,
+      "required_pages_per_second_upper_estimate": 0.0021832100346355723
     },
     {
       "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": 28211.445222682683,
-      "required_pages_per_second_upper_estimate": 0.35029173203226693
+      "required_token_positions_per_second": 28359.723693113316,
+      "required_pages_per_second_upper_estimate": 0.3521328543788981
     }
   ],
   "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": 690670,
-    "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,
+    "producer_batch_seconds": 0.07393975349992843,
+    "producer_p95_seconds": 0.19737816499946348,
+    "gpu_update_seconds": 0.9749037155363348,
+    "producer_batches_per_second": 13.524524395404807,
+    "gpu_batches_per_second": 1.0257423210761472,
+    "producer_headroom": 13.18510908394195,
     "sustainable": true,
-    "buffer_seconds": 60.76172228928425,
+    "buffer_seconds": 60.44403036325276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392381.4397826354,
-    "upload_bytes_per_second": 4951636.511839129,
+    "generation_bytes_per_second": 375850.4545317778,
+    "upload_bytes_per_second": 4881272.278587739,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.19592280492112,
+    "estimated_full_shards_per_hour": 80.6487581917286,
     "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": 1.0843116149999332,
-      "bytes": 11777166,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.0721420089994353,
       "bytes": 6211390,
@@ -562,6 +556,12 @@
       "bytes": 829951,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.183389565001562,
+      "bytes": 6316614,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -124,3 +124,4 @@
 {"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}
+{"step": 249800, "loss": 0.5844248123466969, "lr": 0.0003, "grad_norm": 0.15406217257925406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36563913568, "tokens_per_second": 150778.13998014262, "data_wait_s": 8.314998922287486e-06, "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 sha256:6406f906b48099cc336055624c02fae2f9aa1c471aba59ec6b3460ac95fc0a8d
-size 268599
+oid sha256:ece3ecc30a618d79d1a67b2e75afa5c1081c733ba3cfee88dabddd6207c9111e
+size 270717