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": 76750.25091976103,
-      "required_pages_per_second_upper_estimate": 0.005309760443746543
+      "required_token_positions_per_second": 76877.22818556963,
+      "required_pages_per_second_upper_estimate": 0.005318545025623031
     },
     {
       "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": 9593.78136497013,
-      "required_pages_per_second_upper_estimate": 0.004016715916576565
+      "required_token_positions_per_second": 9609.653523196204,
+      "required_pages_per_second_upper_estimate": 0.004023361257777469
     },
     {
       "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": 19187.56272994026,
-      "required_pages_per_second_upper_estimate": 0.003996263306589665
+      "required_token_positions_per_second": 19219.30704639241,
+      "required_pages_per_second_upper_estimate": 0.0040028748105527265
     },
     {
       "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": 28781.34409491039,
-      "required_pages_per_second_upper_estimate": 0.0017253976319220338
+      "required_token_positions_per_second": 28828.960569588613,
+      "required_pages_per_second_upper_estimate": 0.0017282521668728463
     },
     {
       "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": 28781.34409491039,
-      "required_pages_per_second_upper_estimate": 0.002237622907625661
+      "required_token_positions_per_second": 28828.960569588613,
+      "required_pages_per_second_upper_estimate": 0.0022413248790891555
     },
     {
       "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": 28781.34409491039,
-      "required_pages_per_second_upper_estimate": 0.36917617904991457
+      "required_token_positions_per_second": 28828.960569588613,
+      "required_pages_per_second_upper_estimate": 0.36978695206049966
     }
   ],
   "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": 0.10661142100013876,
-    "producer_p95_seconds": 0.20520348400168587,
-    "gpu_update_seconds": 0.9606222665914061,
-    "producer_batches_per_second": 9.37985809230231,
-    "gpu_batches_per_second": 1.04099190158096,
-    "producer_headroom": 9.010500540933188,
+    "producer_batch_seconds": 0.05886500534376182,
+    "producer_p95_seconds": 0.2048409470007755,
+    "gpu_update_seconds": 0.9590356174396937,
+    "producer_batches_per_second": 16.988021901300556,
+    "gpu_batches_per_second": 1.0427141409718104,
+    "producer_headroom": 16.292118073192817,
     "sustainable": true,
-    "buffer_seconds": 60.519202795258586,
+    "buffer_seconds": 60.4192438987007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328793.51859020645,
-    "upload_bytes_per_second": 5508014.305835382,
+    "generation_bytes_per_second": 353165.97552269616,
+    "upload_bytes_per_second": 5439955.8879699325,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.55143516807217,
+    "estimated_full_shards_per_hour": 75.78119706402458,
     "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": 0.9546346269999049,
-      "bytes": 5999756,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1835868019988993,
       "bytes": 7840439,
@@ -562,6 +556,12 @@
       "bytes": 6241760,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.262870265996753,
+      "bytes": 2380360,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -166,3 +166,4 @@
 {"step": 541600, "loss": 0.5914584994316101, "lr": 0.0003, "grad_norm": 0.16577160816929673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79272297585, "tokens_per_second": 152755.2319532559, "data_wait_s": 1.001299824565649e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541700, "loss": 0.5614579916000366, "lr": 0.0003, "grad_norm": 0.17160825678431055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79287086550, "tokens_per_second": 154765.4882797034, "data_wait_s": 1.0728002962423488e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 541800, "loss": 0.58469158411026, "lr": 0.0003, "grad_norm": 0.15650526434481732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79301820571, "tokens_per_second": 154227.4324284023, "data_wait_s": 1.281200093217194e-05, "gpu_peak_gb": 89.37899398803711}
+{"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}

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:d249ac52c3796f24a78f87a8d8d5ac0639b1804f25c4d3ba16d90b0abefa9a4f
-size 358338
+oid sha256:8fb9601eb3e52d49d8d278cd26ca73bbec34bff864790c74e51df87becf8d190
+size 360456