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": 75741.60551445761,
-      "required_pages_per_second_upper_estimate": 0.006565090400649042
+      "required_token_positions_per_second": 75200.5040346621,
+      "required_pages_per_second_upper_estimate": 0.006518189095789529
     },
     {
       "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": 9467.700689307201,
-      "required_pages_per_second_upper_estimate": 0.003776807359704484
+      "required_token_positions_per_second": 9400.063004332762,
+      "required_pages_per_second_upper_estimate": 0.0037498256758946714
     },
     {
       "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": 18935.401378614402,
-      "required_pages_per_second_upper_estimate": 0.004380371857260136
+      "required_token_positions_per_second": 18800.126008665524,
+      "required_pages_per_second_upper_estimate": 0.004349078281187662
     },
     {
       "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": 28403.1020679216,
-      "required_pages_per_second_upper_estimate": 0.0017295998963767136
+      "required_token_positions_per_second": 28200.189012998286,
+      "required_pages_per_second_upper_estimate": 0.001717243555934406
     },
     {
       "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": 28403.1020679216,
-      "required_pages_per_second_upper_estimate": 0.0021865494220073413
+      "required_token_positions_per_second": 28200.189012998286,
+      "required_pages_per_second_upper_estimate": 0.002170928613340058
     },
     {
       "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": 28403.1020679216,
-      "required_pages_per_second_upper_estimate": 0.35267146861593557
+      "required_token_positions_per_second": 28200.189012998286,
+      "required_pages_per_second_upper_estimate": 0.35015196758009715
     }
   ],
   "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": 0.07606199983587203,
-    "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,
+    "producer_batch_seconds": 0.10955434274211484,
+    "producer_p95_seconds": 0.2733042840009148,
+    "gpu_update_seconds": 0.9804189605699534,
+    "producer_batches_per_second": 9.127890095182693,
+    "gpu_batches_per_second": 1.0199721141854126,
+    "producer_headroom": 8.949156519315789,
     "sustainable": true,
-    "buffer_seconds": 60.3517177771926,
+    "buffer_seconds": 60.78597555533711,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338680.4090955914,
-    "upload_bytes_per_second": 6415526.42953239,
+    "generation_bytes_per_second": 379390.79386084405,
+    "upload_bytes_per_second": 6276862.493143976,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.67293171549612,
+    "estimated_full_shards_per_hour": 81.40843259686461,
     "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": 1.1045795680001902,
-      "bytes": 11876139,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0963092559995857,
       "bytes": 12324311,
@@ -562,6 +556,12 @@
       "bytes": 5885831,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.8101317319997179,
+      "bytes": 713378,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -172,3 +172,4 @@
 {"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}
+{"step": 57500, "loss": 0.6358325146138668, "lr": 0.0003, "grad_norm": 0.129646001522135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8416017932, "tokens_per_second": 150498.3857282842, "data_wait_s": 8.213999535655603e-06, "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 sha256:3b1cce2fbdc940c06c68dce7f39d08a34f37e2a11e8fc514b69713d967e88b1a
-size 371046
+oid sha256:a5a037046f79bb06f590bf3ebc41552a524726f738c240367fd23f94ac060afb
+size 373164