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": 76711.59737064948,
-      "required_pages_per_second_upper_estimate": 0.005307086301530416
+      "required_token_positions_per_second": 77334.85143249403,
+      "required_pages_per_second_upper_estimate": 0.005350204463677485
     },
     {
       "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": 9588.949671331186,
-      "required_pages_per_second_upper_estimate": 0.004014692987347176
+      "required_token_positions_per_second": 9666.856429061754,
+      "required_pages_per_second_upper_estimate": 0.004047310919930865
     },
     {
       "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": 19177.89934266237,
-      "required_pages_per_second_upper_estimate": 0.00399425067786034
+      "required_token_positions_per_second": 19333.71285812351,
+      "required_pages_per_second_upper_estimate": 0.004026702524052169
     },
     {
       "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": 28766.849013993553,
-      "required_pages_per_second_upper_estimate": 0.0017245286739537764
+      "required_token_positions_per_second": 29000.56928718526,
+      "required_pages_per_second_upper_estimate": 0.0017385398474614256
     },
     {
       "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": 28766.849013993553,
-      "required_pages_per_second_upper_estimate": 0.0022364959788415
+      "required_token_positions_per_second": 29000.56928718526,
+      "required_pages_per_second_upper_estimate": 0.002254666702055318
     },
     {
       "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": 28766.849013993553,
-      "required_pages_per_second_upper_estimate": 0.3689902517155187
+      "required_token_positions_per_second": 29000.56928718526,
+      "required_pages_per_second_upper_estimate": 0.37198816443074434
     }
   ],
   "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": 0.08986039678944735,
-    "producer_p95_seconds": 0.322078203993442,
-    "gpu_update_seconds": 0.9611063063094156,
-    "producer_batches_per_second": 11.128372850868981,
-    "gpu_batches_per_second": 1.0404676292677066,
-    "producer_headroom": 10.695549325932667,
+    "producer_batch_seconds": 0.10632708937521329,
+    "producer_p95_seconds": 0.32807919399783714,
+    "gpu_update_seconds": 0.9533605953113847,
+    "producer_batches_per_second": 9.404940978598042,
+    "gpu_batches_per_second": 1.0489210535006244,
+    "producer_headroom": 8.966300130224667,
     "sustainable": true,
-    "buffer_seconds": 60.54969729749318,
+    "buffer_seconds": 60.061717504617235,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300321.8117919923,
-    "upload_bytes_per_second": 5117050.572813109,
+    "generation_bytes_per_second": 329689.041779404,
+    "upload_bytes_per_second": 5121752.382386376,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.4420696765883,
+    "estimated_full_shards_per_hour": 70.74359359776105,
     "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": 1.6674628689979727,
-      "bytes": 7856605,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.3078784290009935,
       "bytes": 11982756,
@@ -562,6 +556,12 @@
       "bytes": 6228637,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2593115519994171,
+      "bytes": 6168643,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -382,3 +382,4 @@
 {"step": 963200, "loss": 0.49379172921180725, "lr": 3e-05, "grad_norm": 0.34721966692841605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140986246013, "tokens_per_second": 151810.60774074096, "data_wait_s": 7.378999725915492e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 963300, "loss": 0.4876033067703247, "lr": 3e-05, "grad_norm": 0.3265877991015855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141000884630, "tokens_per_second": 152203.69996306024, "data_wait_s": 8.148002962116152e-06, "gpu_peak_gb": 89.37899398803711}
 {"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}

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:f1366f336aa5916eee1534acf69744db6c27624f533d800ee10e8b775d8b50fe
-size 818958
+oid sha256:7751baa7345871890b9a1de1f83795275de8fd65151b973234475dff6cf7950e
+size 821076