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": 75625.92984830218,
-      "required_pages_per_second_upper_estimate": 0.006555063927083985
+      "required_token_positions_per_second": 75700.79871476068,
+      "required_pages_per_second_upper_estimate": 0.0065615533706752
     },
     {
       "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": 9453.241231037773,
-      "required_pages_per_second_upper_estimate": 0.003771039265612643
+      "required_token_positions_per_second": 9462.599839345085,
+      "required_pages_per_second_upper_estimate": 0.003774772554390093
     },
     {
       "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": 18906.482462075546,
-      "required_pages_per_second_upper_estimate": 0.004373681974874421
+      "required_token_positions_per_second": 18925.19967869017,
+      "required_pages_per_second_upper_estimate": 0.004378011873526452
     },
     {
       "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": 28359.723693113316,
-      "required_pages_per_second_upper_estimate": 0.0017269583809396325
+      "required_token_positions_per_second": 28387.799518035255,
+      "required_pages_per_second_upper_estimate": 0.0017286680513749096
     },
     {
       "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": 28359.723693113316,
-      "required_pages_per_second_upper_estimate": 0.0021832100346355723
+      "required_token_positions_per_second": 28387.799518035255,
+      "required_pages_per_second_upper_estimate": 0.002185371389356921
     },
     {
       "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": 28359.723693113316,
-      "required_pages_per_second_upper_estimate": 0.3521328543788981
+      "required_token_positions_per_second": 28387.799518035255,
+      "required_pages_per_second_upper_estimate": 0.3524814621606871
     }
   ],
   "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": 0.07393975349992843,
-    "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,
+    "producer_batch_seconds": 0.25185649265637267,
+    "producer_p95_seconds": 0.23409715699926892,
+    "gpu_update_seconds": 0.9739395257612254,
+    "producer_batches_per_second": 3.9705150717094178,
+    "gpu_batches_per_second": 1.026757795067826,
+    "producer_headroom": 3.8670415659684685,
     "sustainable": true,
-    "buffer_seconds": 60.44403036325276,
+    "buffer_seconds": 60.38425059719598,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375850.4545317778,
-    "upload_bytes_per_second": 4881272.278587739,
+    "generation_bytes_per_second": 383732.5768872148,
+    "upload_bytes_per_second": 4854486.8576622605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.6487581917286,
+    "estimated_full_shards_per_hour": 82.34007816278775,
     "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": 1.0721420089994353,
-      "bytes": 6211390,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.7338976749997528,
       "bytes": 1218142,
@@ -562,6 +556,12 @@
       "bytes": 6316614,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.805858101000922,
+      "bytes": 7507792,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -125,3 +125,4 @@
 {"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}
+{"step": 249900, "loss": 0.5893625961616635, "lr": 0.0003, "grad_norm": 0.16091751731171505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36578524362, "tokens_per_second": 149190.4559398415, "data_wait_s": 1.704299938865006e-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:ece3ecc30a618d79d1a67b2e75afa5c1081c733ba3cfee88dabddd6207c9111e
-size 270717
+oid sha256:c00e06ff33c7d3f6c30e3ad12ccd1e81db9cb14b58b81f2b50af86bf7a74a1ee
+size 272835