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": 75500.786373434,
-      "required_pages_per_second_upper_estimate": 0.006544216807855649
+      "required_token_positions_per_second": 74631.11147787124,
+      "required_pages_per_second_upper_estimate": 0.006468835592079139
     },
     {
       "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": 9437.59829667925,
-      "required_pages_per_second_upper_estimate": 0.0037647990652143175
+      "required_token_positions_per_second": 9328.888934733905,
+      "required_pages_per_second_upper_estimate": 0.003721433275384516
     },
     {
       "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": 18875.1965933585,
-      "required_pages_per_second_upper_estimate": 0.004366444539759215
+      "required_token_positions_per_second": 18657.77786946781,
+      "required_pages_per_second_upper_estimate": 0.004316148544425955
     },
     {
       "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": 28312.79489003775,
-      "required_pages_per_second_upper_estimate": 0.0017241006630487317
+      "required_token_positions_per_second": 27986.666804201715,
+      "required_pages_per_second_upper_estimate": 0.0017042411736831458
     },
     {
       "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": 28312.79489003775,
-      "required_pages_per_second_upper_estimate": 0.002179597325467574
+      "required_token_positions_per_second": 27986.666804201715,
+      "required_pages_per_second_upper_estimate": 0.002154491082639594
     },
     {
       "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": 28312.79489003775,
-      "required_pages_per_second_upper_estimate": 0.3515501557053
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27986.666804201715,
+      "required_pages_per_second_upper_estimate": 0.3475007363597069
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.129779342882955,
-    "producer_p95_seconds": 0.35390149500017287,
-    "gpu_update_seconds": 0.9765196303431115,
-    "producer_batches_per_second": 7.705386525973374,
-    "gpu_batches_per_second": 1.0240449540667589,
-    "producer_headroom": 7.524461201994312,
+    "producer_batch_seconds": 0.10867002000782122,
+    "producer_p95_seconds": 0.3626553030044306,
+    "gpu_update_seconds": 0.9878989946687445,
+    "producer_batches_per_second": 9.202170018262882,
+    "gpu_batches_per_second": 1.0122492333695643,
+    "producer_headroom": 9.090814509812764,
     "sustainable": true,
-    "buffer_seconds": 60.54421708127291,
+    "buffer_seconds": 60.26183867479342,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273099.11387248244,
-    "upload_bytes_per_second": 5671831.104403686,
+    "generation_bytes_per_second": 304604.0462717515,
+    "upload_bytes_per_second": 5638935.744139431,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.60071241503577,
+    "estimated_full_shards_per_hour": 65.36093751062783,
     "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": 1.3580094259996258,
-      "bytes": 7460093,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.418776655002148,
       "bytes": 10393041,
@@ -562,6 +556,12 @@
       "bytes": 12063791,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.9416690770012792,
+      "bytes": 2624850,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 305500, "loss": 0.584441065788269, "lr": 0.0003, "grad_norm": 0.18012175779245443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44717283910, "tokens_per_second": 139829.5428343176, "data_wait_s": 1.553699985379353e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 305600, "loss": 0.5479111261665821, "lr": 0.0003, "grad_norm": 0.1525635898032803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44731776385, "tokens_per_second": 138816.0164328992, "data_wait_s": 1.2979995517525822e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 305700, "loss": 0.6008859239518642, "lr": 0.0003, "grad_norm": 0.15546464373632182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44746339856, "tokens_per_second": 149502.92204274362, "data_wait_s": 1.3785996998194605e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 305800, "loss": 0.5321316346526146, "lr": 0.0003, "grad_norm": 0.14079146194573894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44760954425, "tokens_per_second": 149630.37791913783, "data_wait_s": 8.137001714203507e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b51d90d0b7be712ba0a090a3e2c81ff54f947200142c8447ca6fe529e3c7cb39
-size 544170
+oid sha256:9919526e8335e3a72201b6ba19d3a3fab3711bd8fb7a14dc2d6ab6702ed9e803
+size 546288