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": 77073.31812228786,
-      "required_pages_per_second_upper_estimate": 0.005332110982436537
+      "required_token_positions_per_second": 76697.24439996477,
+      "required_pages_per_second_upper_estimate": 0.005306093329714967
     },
     {
       "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": 9634.164765285983,
-      "required_pages_per_second_upper_estimate": 0.004033623602987519
+      "required_token_positions_per_second": 9587.155549995596,
+      "required_pages_per_second_upper_estimate": 0.0040139418262094785
     },
     {
       "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": 19268.329530571966,
-      "required_pages_per_second_upper_estimate": 0.004013084901197483
+      "required_token_positions_per_second": 19174.31109999119,
+      "required_pages_per_second_upper_estimate": 0.003993503341540257
     },
     {
       "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": 28902.49429585795,
-      "required_pages_per_second_upper_estimate": 0.0017326604014831941
+      "required_token_positions_per_second": 28761.46664998679,
+      "required_pages_per_second_upper_estimate": 0.001724206009449965
     },
     {
       "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": 28902.49429585795,
-      "required_pages_per_second_upper_estimate": 0.0022470418028659165
+      "required_token_positions_per_second": 28761.46664998679,
+      "required_pages_per_second_upper_estimate": 0.002236077523714491
     },
     {
       "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": 28902.49429585795,
-      "required_pages_per_second_upper_estimate": 0.37073016374671885
+      "required_token_positions_per_second": 28761.46664998679,
+      "required_pages_per_second_upper_estimate": 0.36892121252917215
     }
   ],
   "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": 0.1161936969845101,
-    "producer_p95_seconds": 0.250281136999547,
-    "gpu_update_seconds": 0.9565956390124526,
-    "producer_batches_per_second": 8.60631881033367,
-    "gpu_batches_per_second": 1.0453737809555104,
-    "producer_headroom": 8.232767041916029,
+    "producer_batch_seconds": 0.10208775592192865,
+    "producer_p95_seconds": 0.28060754499892937,
+    "gpu_update_seconds": 0.9612861658434481,
+    "producer_batches_per_second": 9.795493994056912,
+    "gpu_batches_per_second": 1.0402729546436194,
+    "producer_headroom": 9.416272864089493,
     "sustainable": true,
-    "buffer_seconds": 60.26552525778451,
+    "buffer_seconds": 60.56102844813723,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 171,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 479358.5461322608,
-    "upload_bytes_per_second": 5466962.372045445,
+    "generation_bytes_per_second": 613340.6262448884,
+    "upload_bytes_per_second": 5496470.011946467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.85918510413998,
+    "estimated_full_shards_per_hour": 65.80431027655597,
     "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": 1.2464688429990929,
-      "bytes": 6118587,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8108600380001008,
       "bytes": 790179,
@@ -562,6 +556,12 @@
       "bytes": 1095644,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.0743960049985617,
+      "bytes": 7509044,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -231,3 +231,4 @@
 {"step": 548100, "loss": 0.5568685531616211, "lr": 0.0003, "grad_norm": 0.18329008408751127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80224534705, "tokens_per_second": 153233.11600300542, "data_wait_s": 1.5513000107603148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 548200, "loss": 0.5650813579559326, "lr": 0.0003, "grad_norm": 0.1473897733160872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80239193557, "tokens_per_second": 152586.16318734127, "data_wait_s": 1.5380999684566632e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 548300, "loss": 0.5791099667549133, "lr": 0.0003, "grad_norm": 0.13916699138357838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80253868902, "tokens_per_second": 152728.16984479965, "data_wait_s": 1.3335000403458253e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 548400, "loss": 0.5263413190841675, "lr": 0.0003, "grad_norm": 0.17751341907876914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80268410035, "tokens_per_second": 151864.66621336283, "data_wait_s": 1.2276999768801033e-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:c0faa255ad87020c3bae1e2ee2279c7ef8044c26cb1043d42c130b49618fab31
-size 496791
+oid sha256:3d2bc10aed8d694446cc7ef81879ddfbd526c6af405fdcfcba279629ac6b4eb6
+size 498909