Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76299.85449539992,
-      "required_pages_per_second_upper_estimate": 0.005278600974045556
+      "required_token_positions_per_second": 77020.26655152599,
+      "required_pages_per_second_upper_estimate": 0.0053284407516746264
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9537.48181192499,
-      "required_pages_per_second_upper_estimate": 0.003993144469384923
+      "required_token_positions_per_second": 9627.533318940748,
+      "required_pages_per_second_upper_estimate": 0.0040308471548831165
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19074.96362384998,
-      "required_pages_per_second_upper_estimate": 0.003972811882229174
+      "required_token_positions_per_second": 19255.066637881497,
+      "required_pages_per_second_upper_estimate": 0.004010322590416059
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28612.445435774975,
-      "required_pages_per_second_upper_estimate": 0.0017152724152001855
+      "required_token_positions_per_second": 28882.599956822247,
+      "required_pages_per_second_upper_estimate": 0.0017314677662349006
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28612.445435774975,
-      "required_pages_per_second_upper_estimate": 0.002224491779784571
+      "required_token_positions_per_second": 28882.599956822247,
+      "required_pages_per_second_upper_estimate": 0.0022454951055118403
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28612.445435774975,
-      "required_pages_per_second_upper_estimate": 0.36700972839977647
+      "required_token_positions_per_second": 28882.599956822247,
+      "required_pages_per_second_upper_estimate": 0.3704749805264459
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09201112900002784,
-    "producer_p95_seconds": 0.20955274500011,
-    "gpu_update_seconds": 0.966292799476374,
-    "producer_batches_per_second": 10.868250513475358,
-    "gpu_batches_per_second": 1.0348830091064443,
-    "producer_headroom": 10.501912214076643,
+    "producer_batch_seconds": 0.3428685909374991,
+    "producer_p95_seconds": 0.21483353999974497,
+    "gpu_update_seconds": 0.9572545422272268,
+    "producer_batches_per_second": 2.916569281734786,
+    "gpu_batches_per_second": 1.0446542229753417,
+    "producer_headroom": 2.7918991926610244,
     "sustainable": true,
-    "buffer_seconds": 60.87644636701157,
+    "buffer_seconds": 60.30703616031529,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 173,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 631499.0001856502,
-    "upload_bytes_per_second": 2163042.7768235332,
+    "generation_bytes_per_second": 568762.2773844245,
+    "upload_bytes_per_second": 2140873.359111873,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.75249244774403,
+    "estimated_full_shards_per_hour": 61.02157230925346,
     "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": 1.5430965340001421,
-      "bytes": 6138495,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.151366422999672,
       "bytes": 1194300,
@@ -562,6 +556,12 @@
       "bytes": 13775084,
       "pages": 72,
       "shards": 3
+    },
+    {
+      "seconds": 1.2821505469983094,
+      "bytes": 1156096,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 658800, "loss": 0.5704889893531799, "lr": 0.0001, "grad_norm": 0.19855481073692224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96430477585, "tokens_per_second": 153797.17199487513, "data_wait_s": 1.0458999895490706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658900, "loss": 0.5716390013694763, "lr": 0.0001, "grad_norm": 0.19920145486725915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96445199974, "tokens_per_second": 153532.37566865486, "data_wait_s": 1.3941998986410908e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 659000, "loss": 0.5488200783729553, "lr": 0.0001, "grad_norm": 0.17847421259339394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96459949323, "tokens_per_second": 154457.25089723495, "data_wait_s": 1.0686000678106211e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 659100, "loss": 0.5965697169303894, "lr": 0.0001, "grad_norm": 0.2022988472183881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96474692608, "tokens_per_second": 153723.58693011728, "data_wait_s": 1.2321999747655354e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:044b8674248b215ffe78f4645b6001c6b0f025c6eb2c880c011c3ac857913a23
-size 298251
+oid sha256:c04c795a81cba5c550c922430d71b1514b2942c6869a199da9e82cfd466544a6
+size 300369