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": 75734.38999708506,
-      "required_pages_per_second_upper_estimate": 0.006564464977890751
+      "required_token_positions_per_second": 73362.0431306241,
+      "required_pages_per_second_upper_estimate": 0.006358835964164077
     },
     {
       "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": 9466.798749635633,
-      "required_pages_per_second_upper_estimate": 0.0037764475624842957
+      "required_token_positions_per_second": 9170.255391328012,
+      "required_pages_per_second_upper_estimate": 0.003658151983137072
     },
     {
       "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": 18933.597499271265,
-      "required_pages_per_second_upper_estimate": 0.004379954561521294
+      "required_token_positions_per_second": 18340.510782656023,
+      "required_pages_per_second_upper_estimate": 0.004242754387602069
     },
     {
       "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": 28400.396248906898,
-      "required_pages_per_second_upper_estimate": 0.0017294351261950534
+      "required_token_positions_per_second": 27510.766173984037,
+      "required_pages_per_second_upper_estimate": 0.0016752613221605287
     },
     {
       "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": 28400.396248906898,
-      "required_pages_per_second_upper_estimate": 0.002186341120569403
+      "required_token_positions_per_second": 27510.766173984037,
+      "required_pages_per_second_upper_estimate": 0.0021178549347481766
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28400.396248906898,
-      "required_pages_per_second_upper_estimate": 0.35263787139956665
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27510.766173984037,
+      "required_pages_per_second_upper_estimate": 0.3415916432693549
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4160873022421754,
-    "producer_p95_seconds": 0.19155227099963668,
-    "gpu_update_seconds": 0.9735075439683044,
-    "producer_batches_per_second": 2.4033417857533412,
-    "gpu_batches_per_second": 1.0272134059934497,
-    "producer_headroom": 2.339671359165134,
+    "producer_batch_seconds": 0.059077428218785144,
+    "producer_p95_seconds": 0.19717876499998965,
+    "gpu_update_seconds": 1.0049883680137466,
+    "producer_batches_per_second": 16.926938598217873,
+    "gpu_batches_per_second": 0.9950363922881957,
+    "producer_headroom": 17.011376397291876,
     "sustainable": true,
-    "buffer_seconds": 60.357467726034876,
+    "buffer_seconds": 60.299302080824795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354788.4001987598,
-    "upload_bytes_per_second": 5848753.458101752,
+    "generation_bytes_per_second": 345717.5096777684,
+    "upload_bytes_per_second": 6086515.295810156,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.12933163139435,
+    "estimated_full_shards_per_hour": 74.1829296851138,
     "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": 2283760,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.149097012999846,
+      "bytes": 11248733,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 121700, "loss": 0.6149105243384838, "lr": 0.0003, "grad_norm": 0.13310665873552835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17809178522, "tokens_per_second": 148013.39882786298, "data_wait_s": 1.217600038216915e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 121800, "loss": 0.5454456433653831, "lr": 0.0003, "grad_norm": 0.12661693401183088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17823906849, "tokens_per_second": 150687.54665314627, "data_wait_s": 1.2970000170753337e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 121900, "loss": 0.5915498603135347, "lr": 0.0003, "grad_norm": 0.13542607822741262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17838578756, "tokens_per_second": 148788.3239243314, "data_wait_s": 1.2260000403330196e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 122000, "loss": 0.5904188994318247, "lr": 0.0003, "grad_norm": 0.1396744916789736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17853219445, "tokens_per_second": 149176.60773455244, "data_wait_s": 2.1094000658195e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e27a9ed4a6aa80a201f8d221f8a85a973e07e86d075184a2116ba93b4c459aa
-size 147090
+oid sha256:84177c08b29cc7bab5344a97570d4219cca7f523d3fe675243e2ca9bf6fdf6f1
+size 149208