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": 76382.58516827233,
-      "required_pages_per_second_upper_estimate": 0.0052843244739564765
+      "required_token_positions_per_second": 77277.88931578796,
+      "required_pages_per_second_upper_estimate": 0.0053462636922734395
     },
     {
       "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": 9547.823146034041,
-      "required_pages_per_second_upper_estimate": 0.003997474170024769
+      "required_token_positions_per_second": 9659.736164473496,
+      "required_pages_per_second_upper_estimate": 0.004044329813835794
     },
     {
       "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": 19095.646292068082,
-      "required_pages_per_second_upper_estimate": 0.003977119536580364
+      "required_token_positions_per_second": 19319.47232894699,
+      "required_pages_per_second_upper_estimate": 0.0040237365973727094
     },
     {
       "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": 28643.46943810212,
-      "required_pages_per_second_upper_estimate": 0.0017171322567687893
+      "required_token_positions_per_second": 28979.208493420487,
+      "required_pages_per_second_upper_estimate": 0.0017372593004912709
     },
     {
       "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": 28643.46943810212,
-      "required_pages_per_second_upper_estimate": 0.002226903759505342
+      "required_token_positions_per_second": 28979.208493420487,
+      "required_pages_per_second_upper_estimate": 0.0022530059942963095
     },
     {
       "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": 28643.46943810212,
-      "required_pages_per_second_upper_estimate": 0.36740767099065075
+      "required_token_positions_per_second": 28979.208493420487,
+      "required_pages_per_second_upper_estimate": 0.37171417110376326
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3300422042889295,
-    "producer_p95_seconds": 0.3001118829997722,
-    "gpu_update_seconds": 0.9652461989545886,
-    "producer_batches_per_second": 3.0299155289987336,
-    "gpu_batches_per_second": 1.0360051156720964,
-    "producer_headroom": 2.9246144475195095,
+    "producer_batch_seconds": 0.35064321773418783,
+    "producer_p95_seconds": 0.30760836799890967,
+    "gpu_update_seconds": 0.9540633246169326,
+    "producer_batches_per_second": 2.851901732085034,
+    "gpu_batches_per_second": 1.0481484553465166,
+    "producer_headroom": 2.720894847993836,
     "sustainable": true,
-    "buffer_seconds": 60.810510534139084,
+    "buffer_seconds": 60.10598945086675,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326843.37033808196,
-    "upload_bytes_per_second": 4945314.05569813,
+    "generation_bytes_per_second": 328595.54424433917,
+    "upload_bytes_per_second": 4935504.451971949,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.13297875029416,
+    "estimated_full_shards_per_hour": 70.50895448205596,
     "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": 1.257975387001352,
-      "bytes": 10890364,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.6343959760015423,
       "bytes": 2597952,
@@ -562,6 +556,12 @@
       "bytes": 1158586,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.577455007001845,
+      "bytes": 11622766,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 950700, "loss": 0.5293641686439514, "lr": 3e-05, "grad_norm": 0.33393721874096066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139156385395, "tokens_per_second": 150553.78213193713, "data_wait_s": 9.446001058677211e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 950800, "loss": 0.5309431552886963, "lr": 3e-05, "grad_norm": 0.35004203540961026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139171135510, "tokens_per_second": 153448.81289558203, "data_wait_s": 9.528997907182202e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 950900, "loss": 0.5105096101760864, "lr": 3e-05, "grad_norm": 0.32922813204098184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139185815100, "tokens_per_second": 152203.2716600598, "data_wait_s": 1.1796000762842596e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 951000, "loss": 0.5581480264663696, "lr": 3e-05, "grad_norm": 0.34757897631609186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139200565068, "tokens_per_second": 153829.92213001658, "data_wait_s": 1.072399754775688e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2379c4785b67ddb9fdab2b0b2d90a29f2e4a4b9a64306cb1e8c25520253b6a1d
-size 552642
+oid sha256:3e41118ff91ad079238ec764320eed0dcc273baee3b14c18dee8135bfd8ff296
+size 554760