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": 75976.06598155682,
-      "required_pages_per_second_upper_estimate": 0.005256200533365029
+      "required_token_positions_per_second": 76428.60421528526,
+      "required_pages_per_second_upper_estimate": 0.005287508178407732
     },
     {
       "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": 9497.008247694603,
-      "required_pages_per_second_upper_estimate": 0.00397619903322579
+      "required_token_positions_per_second": 9553.575526910658,
+      "required_pages_per_second_upper_estimate": 0.003999882571774437
     },
     {
       "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": 18994.016495389205,
-      "required_pages_per_second_upper_estimate": 0.003955952730090125
+      "required_token_positions_per_second": 19107.151053821315,
+      "required_pages_per_second_upper_estimate": 0.003979515675052592
     },
     {
       "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": 28491.02474308381,
-      "required_pages_per_second_upper_estimate": 0.001707993430072013
+      "required_token_positions_per_second": 28660.726580731975,
+      "required_pages_per_second_upper_estimate": 0.001718166796119318
     },
     {
       "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": 28491.02474308381,
-      "required_pages_per_second_upper_estimate": 0.0022150518550010175
+      "required_token_positions_per_second": 28660.726580731975,
+      "required_pages_per_second_upper_estimate": 0.002228245426438665
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28491.02474308381,
-      "required_pages_per_second_upper_estimate": 0.3654522741253166
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28660.726580731975,
+      "required_pages_per_second_upper_estimate": 0.3676290270870304
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07112644967186199,
+    "producer_batch_seconds": 0.06535290633589952,
     "producer_p95_seconds": 0.13868397399983223,
-    "gpu_update_seconds": 0.970410866204858,
-    "producer_batches_per_second": 14.059467393823896,
-    "gpu_batches_per_second": 1.030491346321029,
-    "producer_headroom": 13.643459932019603,
+    "gpu_update_seconds": 0.9646650067338903,
+    "producer_batches_per_second": 15.301538310480343,
+    "gpu_batches_per_second": 1.036629288944299,
+    "producer_headroom": 14.7608585573184,
     "sustainable": true,
-    "buffer_seconds": 60.16547370470119,
+    "buffer_seconds": 60.77389542423509,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432985.88275907334,
-    "upload_bytes_per_second": 4151390.5835018563,
+    "generation_bytes_per_second": 434867.1336098362,
+    "upload_bytes_per_second": 4112029.9246091954,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.90869104460859,
+    "estimated_full_shards_per_hour": 93.31236368390384,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 10834593,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8767920650002452,
+      "bytes": 2456869,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 442500, "loss": 0.5862216949462891, "lr": 0.0003, "grad_norm": 0.16555860293504435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64768654750, "tokens_per_second": 151674.82474109088, "data_wait_s": 1.4073999864194775e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442600, "loss": 0.5788451433181763, "lr": 0.0003, "grad_norm": 0.15982561989572341, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64783343283, "tokens_per_second": 153029.0152602861, "data_wait_s": 1.477799969507032e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442700, "loss": 0.5815114378929138, "lr": 0.0003, "grad_norm": 0.15964106845182402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64798010151, "tokens_per_second": 152454.13909953978, "data_wait_s": 1.5870999959588517e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 442800, "loss": 0.5626919269561768, "lr": 0.0003, "grad_norm": 0.16894435588564782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64812683395, "tokens_per_second": 153366.86244665916, "data_wait_s": 1.3889999991079094e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fcdc44027a7d832ea718fdfa83d84b3d48f92a3d3926dd6579b36478b1018203
-size 53115
+oid sha256:2329693d470b6becf3fce62bd73b21ffeb4b562ea0c9b319c79455c52f90b027
+size 55233