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": 77306.99508628547,
-      "required_pages_per_second_upper_estimate": 0.005348277297011149
+      "required_token_positions_per_second": 76561.16853638977,
+      "required_pages_per_second_upper_estimate": 0.005296679285733344
     },
     {
       "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": 9663.374385785684,
-      "required_pages_per_second_upper_estimate": 0.004045853061124507
+      "required_token_positions_per_second": 9570.146067048721,
+      "required_pages_per_second_upper_estimate": 0.004006820310898017
     },
     {
       "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": 19326.74877157137,
-      "required_pages_per_second_upper_estimate": 0.0040252520884786714
+      "required_token_positions_per_second": 19140.292134097443,
+      "required_pages_per_second_upper_estimate": 0.00398641808808505
     },
     {
       "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": 28990.123157357055,
-      "required_pages_per_second_upper_estimate": 0.0017379136179285415
+      "required_token_positions_per_second": 28710.438201146164,
+      "required_pages_per_second_upper_estimate": 0.0017211469318578992
     },
     {
       "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": 28990.123157357055,
-      "required_pages_per_second_upper_estimate": 0.0022538545614088448
+      "required_token_positions_per_second": 28710.438201146164,
+      "required_pages_per_second_upper_estimate": 0.0022321102862675585
     },
     {
       "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": 28990.123157357055,
-      "required_pages_per_second_upper_estimate": 0.37185417269348847
+      "required_token_positions_per_second": 28710.438201146164,
+      "required_pages_per_second_upper_estimate": 0.3682666743775242
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3964736313829178,
-    "producer_p95_seconds": 0.21460396999827935,
-    "gpu_update_seconds": 0.9537041236399006,
-    "producer_batches_per_second": 2.522235833217849,
-    "gpu_batches_per_second": 1.0485432276243147,
-    "producer_headroom": 2.4054667149321833,
+    "producer_batch_seconds": 0.10778374971900462,
+    "producer_p95_seconds": 0.2151292039998225,
+    "gpu_update_seconds": 0.9629947061865551,
+    "producer_batches_per_second": 9.277836432737116,
+    "gpu_batches_per_second": 1.0384273076224741,
+    "producer_headroom": 8.934507369590596,
     "sustainable": true,
-    "buffer_seconds": 60.08335978931374,
+    "buffer_seconds": 60.66866648975297,
     "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": 300474.24916943716,
-    "upload_bytes_per_second": 5432860.288702868,
+    "generation_bytes_per_second": 301357.2234733442,
+    "upload_bytes_per_second": 5483649.158292492,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.47477918922745,
+    "estimated_full_shards_per_hour": 64.66424492025608,
     "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": 1.00542208799925,
-      "bytes": 5360697,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.2967056479992607,
       "bytes": 7612762,
@@ -562,6 +556,12 @@
       "bytes": 5872826,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.268712075001531,
+      "bytes": 10759229,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 541900, "loss": 0.5788364410400391, "lr": 0.0003, "grad_norm": 0.15728552667890988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79316586319, "tokens_per_second": 155135.0005398325, "data_wait_s": 9.436000254936516e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 542000, "loss": 0.5957116484642029, "lr": 0.0003, "grad_norm": 0.151090325777303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79331267739, "tokens_per_second": 152962.60405773998, "data_wait_s": 1.1364998499630019e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 542100, "loss": 0.5550983548164368, "lr": 0.0003, "grad_norm": 0.1649283725345975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79345719644, "tokens_per_second": 150783.55174427692, "data_wait_s": 1.322499883826822e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 542200, "loss": 0.5227675437927246, "lr": 0.0003, "grad_norm": 0.1815667588109992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79360273684, "tokens_per_second": 151991.64493333848, "data_wait_s": 1.5243000234477222e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d23e607d61cecfb29d07c2ca031742011fcb24a02576a1e9632230b3710aec57
-size 364692
+oid sha256:68875b69e7f3bc857c251fcc7ca7d28910ba29635d435d31987be37c2ebbc24d
+size 366810