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": 77313.21059275793,
-      "required_pages_per_second_upper_estimate": 0.005348707300170875
+      "required_token_positions_per_second": 76660.78298839014,
+      "required_pages_per_second_upper_estimate": 0.005303570844660104
     },
     {
       "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": 9664.15132409474,
-      "required_pages_per_second_upper_estimate": 0.004046178348970192
+      "required_token_positions_per_second": 9582.597873548768,
+      "required_pages_per_second_upper_estimate": 0.0040120336222562985
     },
     {
       "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": 19328.30264818948,
-      "required_pages_per_second_upper_estimate": 0.004025575719999742
+      "required_token_positions_per_second": 19165.195747097536,
+      "required_pages_per_second_upper_estimate": 0.003991604853920529
     },
     {
       "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": 28992.453972284224,
-      "required_pages_per_second_upper_estimate": 0.0017380533467244769
+      "required_token_positions_per_second": 28747.793620646302,
+      "required_pages_per_second_upper_estimate": 0.0017233863322185096
     },
     {
       "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": 28992.453972284224,
-      "required_pages_per_second_upper_estimate": 0.0022540357720172606
+      "required_token_positions_per_second": 28747.793620646302,
+      "required_pages_per_second_upper_estimate": 0.0022350145058245684
     },
     {
       "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": 28992.453972284224,
-      "required_pages_per_second_upper_estimate": 0.3718840698847401
+      "required_token_positions_per_second": 28747.793620646302,
+      "required_pages_per_second_upper_estimate": 0.36874582958974744
     }
   ],
   "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": 0.14006221950006648,
-    "producer_p95_seconds": 0.3495665270020254,
-    "gpu_update_seconds": 0.9536274516958974,
-    "producer_batches_per_second": 7.139684088752609,
-    "gpu_batches_per_second": 1.0486275308262523,
-    "producer_headroom": 6.808598743470895,
+    "producer_batch_seconds": 0.375864636625181,
+    "producer_p95_seconds": 0.34481264500209363,
+    "gpu_update_seconds": 0.9617433728946613,
+    "producer_batches_per_second": 2.6605322835870244,
+    "gpu_batches_per_second": 1.0397784150986076,
+    "producer_headroom": 2.5587492921121204,
     "sustainable": true,
-    "buffer_seconds": 60.07852945684154,
+    "buffer_seconds": 60.58983249236366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349333.6393321442,
-    "upload_bytes_per_second": 5121752.382386376,
+    "generation_bytes_per_second": 338857.1891768459,
+    "upload_bytes_per_second": 5004001.579186395,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.95886692975277,
+    "estimated_full_shards_per_hour": 72.71086460570366,
     "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": 1.3078784290009935,
-      "bytes": 11982756,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9895834060007473,
       "bytes": 1128299,
@@ -562,6 +556,12 @@
       "bytes": 6168643,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.411603455999284,
+      "bytes": 2470060,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -384,3 +384,4 @@
 {"step": 963400, "loss": 0.5462270975112915, "lr": 3e-05, "grad_norm": 0.34679611102436675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141015623594, "tokens_per_second": 153549.73405511674, "data_wait_s": 1.4596997061744332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 963500, "loss": 0.5195978283882141, "lr": 3e-05, "grad_norm": 0.34887611637436744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141030197919, "tokens_per_second": 151459.24333926337, "data_wait_s": 1.9879000319633633e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 963600, "loss": 0.5554284453392029, "lr": 3e-05, "grad_norm": 0.33351172778197596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141044863094, "tokens_per_second": 152538.94204790838, "data_wait_s": 1.1431999155320227e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 963700, "loss": 0.524150013923645, "lr": 3e-05, "grad_norm": 0.3474758651429756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141059586896, "tokens_per_second": 152881.51490051608, "data_wait_s": 1.1860996892210096e-05, "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:a278e4b0975eb029f00d7303e21968c7165b7846ab28014c8c91aaca9c25fb5b
-size 823194
+oid sha256:9c0a3f1c782b8741040d708f1e06cf27408fda85fd6c2b9eb576d9ade7888ee7
+size 825312