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": 75362.79979938698,
-      "required_pages_per_second_upper_estimate": 0.006532256481341027
+      "required_token_positions_per_second": 75741.60551445761,
+      "required_pages_per_second_upper_estimate": 0.006565090400649042
     },
     {
       "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": 9420.349974923372,
-      "required_pages_per_second_upper_estimate": 0.003757918451780506
+      "required_token_positions_per_second": 9467.700689307201,
+      "required_pages_per_second_upper_estimate": 0.003776807359704484
     },
     {
       "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": 18840.699949846745,
-      "required_pages_per_second_upper_estimate": 0.004358464348402961
+      "required_token_positions_per_second": 18935.401378614402,
+      "required_pages_per_second_upper_estimate": 0.004380371857260136
     },
     {
       "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": 28261.04992477012,
-      "required_pages_per_second_upper_estimate": 0.0017209496661487844
+      "required_token_positions_per_second": 28403.1020679216,
+      "required_pages_per_second_upper_estimate": 0.0017295998963767136
     },
     {
       "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": 28261.04992477012,
-      "required_pages_per_second_upper_estimate": 0.00217561385480204
+      "required_token_positions_per_second": 28403.1020679216,
+      "required_pages_per_second_upper_estimate": 0.0021865494220073413
     },
     {
       "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": 28261.04992477012,
-      "required_pages_per_second_upper_estimate": 0.3509076564159345
+      "required_token_positions_per_second": 28403.1020679216,
+      "required_pages_per_second_upper_estimate": 0.35267146861593557
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07716415895299633,
-    "producer_p95_seconds": 0.21215325299999677,
-    "gpu_update_seconds": 0.9783076026403111,
-    "producer_batches_per_second": 12.95938442884006,
-    "gpu_batches_per_second": 1.022173391376234,
-    "producer_headroom": 12.678264312272697,
+    "producer_batch_seconds": 0.07606199983587203,
+    "producer_p95_seconds": 0.20588138299717684,
+    "gpu_update_seconds": 0.9734148028579451,
+    "producer_batches_per_second": 13.147169442794276,
+    "gpu_batches_per_second": 1.0273112727112848,
+    "producer_headroom": 12.79764935129759,
     "sustainable": true,
-    "buffer_seconds": 60.65507136369929,
+    "buffer_seconds": 60.3517177771926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352918.2639325382,
-    "upload_bytes_per_second": 6368556.419869212,
+    "generation_bytes_per_second": 338680.4090955914,
+    "upload_bytes_per_second": 6415526.42953239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.72804392320738,
+    "estimated_full_shards_per_hour": 72.67293171549612,
     "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": 0.756838974000857,
-      "bytes": 1060731,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1045795680001902,
       "bytes": 11876139,
@@ -562,6 +556,12 @@
       "bytes": 18261180,
       "pages": 44,
       "shards": 5
+    },
+    {
+      "seconds": 1.0190579750014876,
+      "bytes": 5885831,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -171,3 +171,4 @@
 {"step": 57100, "loss": 0.6742852330207825, "lr": 0.0003, "grad_norm": 0.15001693965512816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8357399409, "tokens_per_second": 149993.12242864975, "data_wait_s": 1.1885000276379287e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57200, "loss": 0.6306101009249687, "lr": 0.0003, "grad_norm": 0.13996838550861912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8371995313, "tokens_per_second": 148911.11273361236, "data_wait_s": 1.0081999789690599e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57300, "loss": 0.600916426628828, "lr": 0.0003, "grad_norm": 0.13766784713913385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8386571728, "tokens_per_second": 149561.33725087225, "data_wait_s": 1.091299782274291e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 57400, "loss": 0.6162293180823326, "lr": 0.0003, "grad_norm": 0.13353620542181208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8401222730, "tokens_per_second": 149766.70442513688, "data_wait_s": 1.0446001397212967e-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 sha256:5b9aaba9e6c492f127babf5d575f3aac0dcebf7c9b024ef7c1a01a128b08cc84
-size 368928
+oid sha256:3b1cce2fbdc940c06c68dce7f39d08a34f37e2a11e8fc514b69713d967e88b1a
+size 371046