Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75808.43273447872,
-      "required_pages_per_second_upper_estimate": 0.00657088281470843
+      "required_token_positions_per_second": 75823.66288437131,
+      "required_pages_per_second_upper_estimate": 0.006572202925500659
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9476.05409180984,
-      "required_pages_per_second_upper_estimate": 0.003780139656857284
+      "required_token_positions_per_second": 9477.957860546414,
+      "required_pages_per_second_upper_estimate": 0.0037808990986701826
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18952.10818361968,
-      "required_pages_per_second_upper_estimate": 0.004384236682568382
+      "required_token_positions_per_second": 18955.915721092828,
+      "required_pages_per_second_upper_estimate": 0.004385117489352424
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28428.162275429517,
-      "required_pages_per_second_upper_estimate": 0.0017311259315331983
+      "required_token_positions_per_second": 28433.873581639244,
+      "required_pages_per_second_upper_estimate": 0.0017314737201164626
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28428.162275429517,
-      "required_pages_per_second_upper_estimate": 0.002188478626152395
+      "required_token_positions_per_second": 28433.873581639244,
+      "required_pages_per_second_upper_estimate": 0.002188918298314327
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28428.162275429517,
-      "required_pages_per_second_upper_estimate": 0.3529826325220646
+      "required_token_positions_per_second": 28433.873581639244,
+      "required_pages_per_second_upper_estimate": 0.3530535478306771
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2652045228905422,
-    "producer_p95_seconds": 0.37430268400203204,
-    "gpu_update_seconds": 0.9725567109167724,
-    "producer_batches_per_second": 3.7706747573559665,
-    "gpu_batches_per_second": 1.0282176748925607,
-    "producer_headroom": 3.6671950399510176,
+    "producer_batch_seconds": 0.11805890318737511,
+    "producer_p95_seconds": 0.38294975900498684,
+    "gpu_update_seconds": 0.9723613604955074,
+    "producer_batches_per_second": 8.470348046626079,
+    "gpu_batches_per_second": 1.0284242470210954,
+    "producer_headroom": 8.236239150487796,
     "sustainable": true,
-    "buffer_seconds": 60.29851607683989,
+    "buffer_seconds": 60.286404350721455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473796.8373570158,
-    "upload_bytes_per_second": 6182898.725273567,
+    "generation_bytes_per_second": 455491.23645555234,
+    "upload_bytes_per_second": 6291518.596442323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.66577187092643,
+    "estimated_full_shards_per_hour": 97.73781605005195,
     "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": 1.1073444919966278,
-      "bytes": 2450943,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 0.8343555079991347,
       "bytes": 5999582,
@@ -562,6 +556,12 @@
       "bytes": 1132933,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.4302320659990073,
+      "bytes": 11751486,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 33200, "loss": 0.6615162566304207, "lr": 0.0003, "grad_norm": 0.15205180377260413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4858475990, "tokens_per_second": 147517.98259767075, "data_wait_s": 1.0149000445380807e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 33300, "loss": 0.6339767333120108, "lr": 0.0003, "grad_norm": 0.13285349390291137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4872968460, "tokens_per_second": 147820.88975675576, "data_wait_s": 9.909002983476967e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 33400, "loss": 0.662850197404623, "lr": 0.0003, "grad_norm": 0.15025443419744325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4887548096, "tokens_per_second": 149562.6214300082, "data_wait_s": 1.2087999493815005e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 33500, "loss": 0.6309784017503262, "lr": 0.0003, "grad_norm": 0.12907676687920205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4902127146, "tokens_per_second": 148909.81059540564, "data_wait_s": 1.2658005289267749e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9caa50877ffac93b2be1224df290888356febdb47e5b6a1733c65f8a08145a7
-size 706501
+oid sha256:ef17ea013ec959e32e08f0408fc76cd151435786b997e30ec3f8df6012660aeb
+size 708619