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": 76259.90825421158,
-      "required_pages_per_second_upper_estimate": 0.0052758374003920964
+      "required_token_positions_per_second": 76872.99954487137,
+      "required_pages_per_second_upper_estimate": 0.005318252478447727
     },
     {
       "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": 9532.488531776447,
-      "required_pages_per_second_upper_estimate": 0.003991053887258274
+      "required_token_positions_per_second": 9609.124943108922,
+      "required_pages_per_second_upper_estimate": 0.004023139952332931
     },
     {
       "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": 19064.977063552895,
-      "required_pages_per_second_upper_estimate": 0.003970731945082596
+      "required_token_positions_per_second": 19218.249886217844,
+      "required_pages_per_second_upper_estimate": 0.004002654631967553
     },
     {
       "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": 28597.465595329344,
-      "required_pages_per_second_upper_estimate": 0.001714374396638362
+      "required_token_positions_per_second": 28827.374829326767,
+      "required_pages_per_second_upper_estimate": 0.0017281571041654338
     },
     {
       "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": 28597.465595329344,
-      "required_pages_per_second_upper_estimate": 0.002223327163079278
+      "required_token_positions_per_second": 28827.374829326767,
+      "required_pages_per_second_upper_estimate": 0.002241201594758731
     },
     {
       "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": 28597.465595329344,
-      "required_pages_per_second_upper_estimate": 0.3668175830906395
+      "required_token_positions_per_second": 28827.374829326767,
+      "required_pages_per_second_upper_estimate": 0.36976661188705595
     }
   ],
   "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": 0.1262510905002614,
-    "producer_p95_seconds": 0.35360358999605523,
-    "gpu_update_seconds": 0.9667989601328723,
-    "producer_batches_per_second": 7.920723662960595,
-    "gpu_batches_per_second": 1.0343412035347708,
-    "producer_headroom": 7.657747400850138,
+    "producer_batch_seconds": 0.5551082932656755,
+    "producer_p95_seconds": 0.35018453399970895,
+    "gpu_update_seconds": 0.9590883722049168,
+    "producer_batches_per_second": 1.8014502974132272,
+    "gpu_batches_per_second": 1.0426567863616452,
+    "producer_headroom": 1.7277500333541154,
     "sustainable": true,
-    "buffer_seconds": 60.908334488370954,
+    "buffer_seconds": 60.42256744890976,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370066.79292079154,
-    "upload_bytes_per_second": 5033287.253275363,
+    "generation_bytes_per_second": 322364.78605369857,
+    "upload_bytes_per_second": 5121692.083898635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.40771904676257,
+    "estimated_full_shards_per_hour": 69.17197881897181,
     "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": 0.9417887669987977,
-      "bytes": 2628206,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.5466363100022136,
       "bytes": 15617223,
@@ -562,6 +556,12 @@
       "bytes": 1290324,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.2432711480068974,
+      "bytes": 11617672,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -403,3 +403,4 @@
 {"step": 965300, "loss": 0.5464724898338318, "lr": 3e-05, "grad_norm": 0.3479924102645416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141293728848, "tokens_per_second": 150881.69473970743, "data_wait_s": 1.3823002518620342e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 965400, "loss": 0.5208156108856201, "lr": 3e-05, "grad_norm": 0.35244522097223124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141308270148, "tokens_per_second": 151777.1182755648, "data_wait_s": 1.0207993909716606e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 965500, "loss": 0.49153828620910645, "lr": 3e-05, "grad_norm": 0.341800847071638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141322690087, "tokens_per_second": 127163.39158951252, "data_wait_s": 1.6079997294582427e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 965600, "loss": 0.5570799708366394, "lr": 3e-05, "grad_norm": 0.3516265639521352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141337341470, "tokens_per_second": 151800.38202151313, "data_wait_s": 9.75099828792736e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f77b6bb295e5a380918517eff8e7a23fee2018bb4cb198937fb28546d18b2184
-size 864219
+oid sha256:682647760f6594da92d633212875084f8a8184b2ed302a25e6374763988bff5c
+size 866337