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": 76060.2369640391,
-      "required_pages_per_second_upper_estimate": 0.00526202367723674
+      "required_token_positions_per_second": 75852.5568213337,
+      "required_pages_per_second_upper_estimate": 0.005247655883080059
     },
     {
       "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": 9507.529620504887,
-      "required_pages_per_second_upper_estimate": 0.003980604112310208
+      "required_token_positions_per_second": 9481.569602666712,
+      "required_pages_per_second_upper_estimate": 0.003969735194948181
     },
     {
       "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": 19015.059241009774,
-      "required_pages_per_second_upper_estimate": 0.00396033537906837
+      "required_token_positions_per_second": 18963.139205333424,
+      "required_pages_per_second_upper_estimate": 0.003949521804860403
     },
     {
       "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": 28522.58886151466,
-      "required_pages_per_second_upper_estimate": 0.0017098856507763245
+      "required_token_positions_per_second": 28444.708808000134,
+      "required_pages_per_second_upper_estimate": 0.0017052168604840847
     },
     {
       "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": 28522.58886151466,
-      "required_pages_per_second_upper_estimate": 0.002217505826373132
+      "required_token_positions_per_second": 28444.708808000134,
+      "required_pages_per_second_upper_estimate": 0.0022114509947705305
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28522.58886151466,
-      "required_pages_per_second_upper_estimate": 0.3658571447456377
+      "required_token_positions_per_second": 28444.708808000134,
+      "required_pages_per_second_upper_estimate": 0.3648581830402398
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12601667296064534,
-    "producer_p95_seconds": 0.3394651169946883,
-    "gpu_update_seconds": 0.969336974783003,
-    "producer_batches_per_second": 7.935457876374004,
-    "gpu_batches_per_second": 1.0316329883360338,
-    "producer_headroom": 7.69213273140233,
+    "producer_batch_seconds": 0.32064328253159147,
+    "producer_p95_seconds": 0.34536966800078517,
+    "gpu_update_seconds": 0.97199096628558,
+    "producer_batches_per_second": 3.118730547244428,
+    "gpu_batches_per_second": 1.0288161461226901,
+    "producer_headroom": 3.031377918200467,
     "sustainable": true,
-    "buffer_seconds": 60.09889243654618,
+    "buffer_seconds": 60.26343990970596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326557.2649149463,
-    "upload_bytes_per_second": 2963081.5292745796,
+    "generation_bytes_per_second": 366620.5608714485,
+    "upload_bytes_per_second": 2942932.417722487,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.07158718668262,
+    "estimated_full_shards_per_hour": 78.66823787314979,
     "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": 1.4824117819953244,
-      "bytes": 7847757,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.698374239000259,
       "bytes": 5912160,
@@ -562,6 +556,12 @@
       "bytes": 2408393,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.9536431870001252,
+      "bytes": 6335446,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -415,3 +415,4 @@
 {"step": 884000, "loss": 0.515288770198822, "lr": 3e-05, "grad_norm": 0.30551710940741256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129392852352, "tokens_per_second": 151967.4666830402, "data_wait_s": 1.2092001270502806e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884100, "loss": 0.6586899757385254, "lr": 3e-05, "grad_norm": 0.607955071258243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129407429533, "tokens_per_second": 151425.7244361318, "data_wait_s": 1.3840995961800218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884200, "loss": 0.507845938205719, "lr": 3e-05, "grad_norm": 0.322197629398531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129421974009, "tokens_per_second": 151447.7346586818, "data_wait_s": 1.6657999367453158e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 884300, "loss": 0.5248423218727112, "lr": 3e-05, "grad_norm": 0.3125030448172315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129436576410, "tokens_per_second": 151040.53643126355, "data_wait_s": 1.5714998880866915e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99d8e1f50d32216ba5284be8bd1e9fee2f19b97c1ba73117fce273f057d8c423
-size 889635
+oid sha256:f3d9319c1bb662d03a40bf0923d6720d70c5957e01dfd1d42f7d4843792b0311
+size 891753