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": 74968.19269653196,
-      "required_pages_per_second_upper_estimate": 0.006498052937788113
+      "required_token_positions_per_second": 75599.79864866537,
+      "required_pages_per_second_upper_estimate": 0.006552798940928386
     },
     {
       "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": 9371.024087066495,
-      "required_pages_per_second_upper_estimate": 0.0037382416176266536
+      "required_token_positions_per_second": 9449.974831083171,
+      "required_pages_per_second_upper_estimate": 0.0037697362498337206
     },
     {
       "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": 18742.04817413299,
-      "required_pages_per_second_upper_estimate": 0.004335642996303537
+      "required_token_positions_per_second": 18899.949662166342,
+      "required_pages_per_second_upper_estimate": 0.004372170726588778
     },
     {
       "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": 28113.072261199482,
-      "required_pages_per_second_upper_estimate": 0.0017119386028161315
+      "required_token_positions_per_second": 28349.924493249513,
+      "required_pages_per_second_upper_estimate": 0.0017263616610803552
     },
     {
       "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": 28113.072261199482,
-      "required_pages_per_second_upper_estimate": 0.0021642221246319826
+      "required_token_positions_per_second": 28349.924493249513,
+      "required_pages_per_second_upper_estimate": 0.0021824556651041356
     },
     {
       "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": 28113.072261199482,
-      "required_pages_per_second_upper_estimate": 0.34907026908376876
+      "required_token_positions_per_second": 28349.924493249513,
+      "required_pages_per_second_upper_estimate": 0.3520111811124019
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06400749672661732,
-    "producer_p95_seconds": 0.168489010000485,
-    "gpu_update_seconds": 0.9834570815712178,
-    "producer_batches_per_second": 15.623169958842542,
-    "gpu_batches_per_second": 1.0168211900028747,
-    "producer_headroom": 15.364717132614409,
+    "producer_batch_seconds": 0.20522467597652394,
+    "producer_p95_seconds": 0.17892680000022665,
+    "gpu_update_seconds": 0.9752406926721569,
+    "producer_batches_per_second": 4.872708387730099,
+    "gpu_batches_per_second": 1.0253878939977399,
+    "producer_headroom": 4.752063503239331,
     "sustainable": true,
-    "buffer_seconds": 60.974339057415506,
+    "buffer_seconds": 60.46492294567373,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 466349.39976008213,
-    "upload_bytes_per_second": 5862724.749249353,
+    "generation_bytes_per_second": 453147.41967767646,
+    "upload_bytes_per_second": 6059835.903003321,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.0677251300988,
+    "estimated_full_shards_per_hour": 97.23488753078193,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 1092998,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1609742079999705,
+      "bytes": 10556650,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -45,3 +45,4 @@
 {"step": 44500, "loss": 0.5941246263682842, "lr": 0.0003, "grad_norm": 0.14556532383493165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6512179341, "tokens_per_second": 148924.71859996233, "data_wait_s": 1.2671999684243929e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 44600, "loss": 0.6809761300683022, "lr": 0.0003, "grad_norm": 0.13601700008440684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6526582608, "tokens_per_second": 146852.09765430127, "data_wait_s": 9.774999853107147e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 44700, "loss": 0.6488265637308359, "lr": 0.0003, "grad_norm": 0.13343255495650386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6541086829, "tokens_per_second": 147309.0517887333, "data_wait_s": 8.866999451129232e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 44800, "loss": 0.6016717441380024, "lr": 0.0003, "grad_norm": 0.1310367573175083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6555559996, "tokens_per_second": 147289.71960988585, "data_wait_s": 1.4893999832565896e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6cbeffafd53b9f2859b3fc829f03c8ffb878d01cbc3d2f28f9758a60e03ce177
-size 99711
+oid sha256:e09abe245f42b0039cb60064fa3eb9e454b78f9cccbe3dfa4037b2ad6bca3bad
+size 101829