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": 76790.61688649825,
-      "required_pages_per_second_upper_estimate": 0.005312553054987371
+      "required_token_positions_per_second": 76977.52696862702,
+      "required_pages_per_second_upper_estimate": 0.0053254839281601804
     },
     {
       "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": 9598.82711081228,
-      "required_pages_per_second_upper_estimate": 0.004018828464993519
+      "required_token_positions_per_second": 9622.190871078377,
+      "required_pages_per_second_upper_estimate": 0.00402861038352614
     },
     {
       "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": 19197.65422162456,
-      "required_pages_per_second_upper_estimate": 0.003998365098176973
+      "required_token_positions_per_second": 19244.381742156755,
+      "required_pages_per_second_upper_estimate": 0.004008097208416245
     },
     {
       "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": 28796.48133243684,
-      "required_pages_per_second_upper_estimate": 0.001726305086198521
+      "required_token_positions_per_second": 28866.572613235134,
+      "required_pages_per_second_upper_estimate": 0.0017305069514591898
     },
     {
       "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": 28796.48133243684,
-      "required_pages_per_second_upper_estimate": 0.0022387997612616714
+      "required_token_positions_per_second": 28866.572613235134,
+      "required_pages_per_second_upper_estimate": 0.0022442490500447763
     },
     {
       "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": 28796.48133243684,
-      "required_pages_per_second_upper_estimate": 0.3693703432798045
+      "required_token_positions_per_second": 28866.572613235134,
+      "required_pages_per_second_upper_estimate": 0.37026939897173117
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09503345663283369,
-    "producer_p95_seconds": 0.18286719099978654,
-    "gpu_update_seconds": 0.9601173032504089,
-    "producer_batches_per_second": 10.522609988433315,
-    "gpu_batches_per_second": 1.0415394000447353,
-    "producer_headroom": 10.10293992525041,
+    "producer_batch_seconds": 0.2382914912343992,
+    "producer_p95_seconds": 0.19726838000042335,
+    "gpu_update_seconds": 0.9577860305910919,
+    "producer_batches_per_second": 4.196540945796231,
+    "gpu_batches_per_second": 1.044074530281942,
+    "producer_headroom": 4.019388294687159,
     "sustainable": true,
-    "buffer_seconds": 60.48739010477576,
+    "buffer_seconds": 60.34051992723879,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 163,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 596176.34629624,
+    "generation_bytes_per_second": 531171.3720406418,
     "upload_bytes_per_second": 2986327.137644053,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.96278282005978,
+    "estimated_full_shards_per_hour": 113.97701140322152,
     "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": [

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 692100, "loss": 0.5424365401268005, "lr": 8e-05, "grad_norm": 0.2043193890694023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101304951523, "tokens_per_second": 153630.61658268468, "data_wait_s": 1.5503999748034403e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692200, "loss": 0.5909315943717957, "lr": 8e-05, "grad_norm": 0.21854433595812459, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101319637046, "tokens_per_second": 153310.22803827678, "data_wait_s": 1.6062000213423744e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692300, "loss": 0.5673353672027588, "lr": 8e-05, "grad_norm": 0.19820353962459786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101334258242, "tokens_per_second": 152745.78443007293, "data_wait_s": 1.7055999705917202e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 692400, "loss": 0.5341150164604187, "lr": 8e-05, "grad_norm": 0.2055547347648032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101348716792, "tokens_per_second": 150329.32750909438, "data_wait_s": 1.503000021330081e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b12774e34c00f645e7d8699dcf3c06bdfc375f5902110538975341e79ffccad
-size 155562
+oid sha256:1d631c25a94d272a69ca485d69018900f33f6fbc616599966f01acf219cc5a27
+size 157680