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": 75901.98374872892,
-      "required_pages_per_second_upper_estimate": 0.0065789915795207606
+      "required_token_positions_per_second": 75803.83688618052,
+      "required_pages_per_second_upper_estimate": 0.006570484458226014
     },
     {
       "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": 9487.747968591115,
-      "required_pages_per_second_upper_estimate": 0.0037848045191443734
+      "required_token_positions_per_second": 9475.479610772565,
+      "required_pages_per_second_upper_estimate": 0.003779910487782258
     },
     {
       "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": 18975.49593718223,
-      "required_pages_per_second_upper_estimate": 0.004389647027744673
+      "required_token_positions_per_second": 18950.95922154513,
+      "required_pages_per_second_upper_estimate": 0.0043839708904134765
     },
     {
       "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": 28463.243905773343,
-      "required_pages_per_second_upper_estimate": 0.0017332622187620475
+      "required_token_positions_per_second": 28426.438832317697,
+      "required_pages_per_second_upper_estimate": 0.0017310209828898956
     },
     {
       "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": 28463.243905773343,
-      "required_pages_per_second_upper_estimate": 0.002191179307168431
+      "required_token_positions_per_second": 28426.438832317697,
+      "required_pages_per_second_upper_estimate": 0.002188345950730851
     },
     {
       "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": 28463.243905773343,
-      "required_pages_per_second_upper_estimate": 0.353418228960271
+      "required_token_positions_per_second": 28426.438832317697,
+      "required_pages_per_second_upper_estimate": 0.3529612331266089
     }
   ],
   "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": 0.09726250359392452,
-    "producer_p95_seconds": 0.2948665549993166,
-    "gpu_update_seconds": 0.9713580114595447,
-    "producer_batches_per_second": 10.281454445950173,
-    "gpu_batches_per_second": 1.0294865417308068,
-    "producer_headroom": 9.986973145530056,
+    "producer_batch_seconds": 0.0878431951641403,
+    "producer_p95_seconds": 0.2905515259990352,
+    "gpu_update_seconds": 0.9726156752553647,
+    "producer_batches_per_second": 11.383921066753546,
+    "gpu_batches_per_second": 1.028155339710565,
+    "producer_headroom": 11.072180075394273,
     "sustainable": true,
-    "buffer_seconds": 60.224196710491775,
+    "buffer_seconds": 60.30217186583261,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307789.8196931588,
-    "upload_bytes_per_second": 5625259.419103601,
+    "generation_bytes_per_second": 289090.7344836702,
+    "upload_bytes_per_second": 5585905.935574231,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.04453032585214,
+    "estimated_full_shards_per_hour": 62.03214193232135,
     "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": 1.1150978799996665,
-      "bytes": 7568611,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 1.1983624430013151,
       "bytes": 7531256,
@@ -562,6 +556,12 @@
       "bytes": 4937209,
       "pages": 2,
       "shards": 2
+    },
+    {
+      "seconds": 1.0983402380006737,
+      "bytes": 4472235,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -199,3 +199,4 @@
 {"step": 19900, "loss": 0.696993101388216, "lr": 0.0003, "grad_norm": 0.1368176895831353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2911812605, "tokens_per_second": 149237.22631955068, "data_wait_s": 1.035999957821332e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20000, "loss": 0.7300440222024918, "lr": 0.0003, "grad_norm": 0.14907453662659742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2926617770, "tokens_per_second": 149959.90069238516, "data_wait_s": 1.1936001101275906e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20100, "loss": 0.6706032436341047, "lr": 0.0003, "grad_norm": 0.13800229756680835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2941240165, "tokens_per_second": 134596.65403390976, "data_wait_s": 1.006900129141286e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 20200, "loss": 0.6517019756138325, "lr": 0.0003, "grad_norm": 0.14619521083157333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2955752885, "tokens_per_second": 124759.274725961, "data_wait_s": 1.484200038248673e-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:fd884d45ce9de0633a4807bd605481e3a697d025096c56e4e01878f882090a64
-size 423241
+oid sha256:bfee234fcd225def567abf0c1bfb8a562c7cb60a611e84dbbac1a7e005cbfdde
+size 425359