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": 77401.15202414618,
-      "required_pages_per_second_upper_estimate": 0.005354791292446549
+      "required_token_positions_per_second": 77135.80967804482,
+      "required_pages_per_second_upper_estimate": 0.00533643429326419
     },
     {
       "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": 9675.144003018273,
-      "required_pages_per_second_upper_estimate": 0.004050780754082235
+      "required_token_positions_per_second": 9641.976209755603,
+      "required_pages_per_second_upper_estimate": 0.00403689409166544
     },
     {
       "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": 19350.288006036546,
-      "required_pages_per_second_upper_estimate": 0.00403015469024641
+      "required_token_positions_per_second": 19283.952419511206,
+      "required_pages_per_second_upper_estimate": 0.00401633873696024
     },
     {
       "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": 29025.432009054817,
-      "required_pages_per_second_upper_estimate": 0.0017400303296743274
+      "required_token_positions_per_second": 28925.928629266808,
+      "required_pages_per_second_upper_estimate": 0.0017340652539889
     },
     {
       "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": 29025.432009054817,
-      "required_pages_per_second_upper_estimate": 0.0022565996693211204
+      "required_token_positions_per_second": 28925.928629266808,
+      "required_pages_per_second_upper_estimate": 0.002248863719211716
     },
     {
       "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": 29025.432009054817,
-      "required_pages_per_second_upper_estimate": 0.37230707673137614
+      "required_token_positions_per_second": 28925.928629266808,
+      "required_pages_per_second_upper_estimate": 0.3710307542138609
     }
   ],
   "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": 0.28477932668755557,
-    "producer_p95_seconds": 0.2997639219975099,
-    "gpu_update_seconds": 0.9525439618392204,
-    "producer_batches_per_second": 3.511490850236983,
-    "gpu_batches_per_second": 1.0498203128275037,
-    "producer_headroom": 3.344849406446908,
+    "producer_batch_seconds": 0.08379916799225384,
+    "producer_p95_seconds": 0.299602525003138,
+    "gpu_update_seconds": 0.9558206533091622,
+    "producer_batches_per_second": 11.933292703961419,
+    "gpu_batches_per_second": 1.0462213769266062,
+    "producer_headroom": 11.406087628429862,
     "sustainable": true,
-    "buffer_seconds": 60.01026959587089,
+    "buffer_seconds": 60.21670115847722,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 410797.2128236449,
-    "upload_bytes_per_second": 4929875.323082814,
+    "generation_bytes_per_second": 411913.9450000528,
+    "upload_bytes_per_second": 4923504.85507691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.14751900226604,
+    "estimated_full_shards_per_hour": 88.38714373112857,
     "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": 1.1741747699998086,
-      "bytes": 5919198,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.178343194998888,
       "bytes": 10619543,
@@ -562,6 +556,12 @@
       "bytes": 6016198,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.177016456000274,
+      "bytes": 5382045,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -333,3 +333,4 @@
 {"step": 590400, "loss": 0.5338912606239319, "lr": 0.0003, "grad_norm": 0.15959345216673101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86416594215, "tokens_per_second": 149993.94368846095, "data_wait_s": 1.1562999134184793e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590500, "loss": 0.6035844683647156, "lr": 0.0003, "grad_norm": 0.16151404817080603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86431196907, "tokens_per_second": 152423.2747475952, "data_wait_s": 1.316200359724462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590600, "loss": 0.5551534295082092, "lr": 0.0003, "grad_norm": 0.16390083626896032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86445968715, "tokens_per_second": 154027.7580806967, "data_wait_s": 8.51499498821795e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 590700, "loss": 0.6077341437339783, "lr": 0.0003, "grad_norm": 0.15848083511875027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86460672697, "tokens_per_second": 153083.06157210996, "data_wait_s": 1.0315998224541545e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8562fb9d1d067506ad1dafc5c308b57abf520782dbe3499ec58b5489bdc0aad8
-size 715176
+oid sha256:6f5976649da0359eab0c1cb9550064b22d1bd24294eb938de4a8fcfbf26c6cff
+size 717294