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": 77155.26209510727,
-      "required_pages_per_second_upper_estimate": 0.005337780056612396
+      "required_token_positions_per_second": 76445.36002471598,
+      "required_pages_per_second_upper_estimate": 0.005288667384182983
     },
     {
       "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": 9644.40776188841,
-      "required_pages_per_second_upper_estimate": 0.004037912131766863
+      "required_token_positions_per_second": 9555.670003089497,
+      "required_pages_per_second_upper_estimate": 0.004000759485212881
     },
     {
       "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": 19288.81552377682,
-      "required_pages_per_second_upper_estimate": 0.004017351593330083
+      "required_token_positions_per_second": 19111.340006178994,
+      "required_pages_per_second_upper_estimate": 0.003980388123358595
     },
     {
       "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": 28933.22328566523,
-      "required_pages_per_second_upper_estimate": 0.00173450255750169
+      "required_token_positions_per_second": 28667.01000926849,
+      "required_pages_per_second_upper_estimate": 0.0017185434780658424
     },
     {
       "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": 28933.22328566523,
-      "required_pages_per_second_upper_estimate": 0.002249430846660891
+      "required_token_positions_per_second": 28667.01000926849,
+      "required_pages_per_second_upper_estimate": 0.0022287339353694977
     },
     {
       "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": 28933.22328566523,
-      "required_pages_per_second_upper_estimate": 0.371124322233748
+      "required_token_positions_per_second": 28667.01000926849,
+      "required_pages_per_second_upper_estimate": 0.3677096241616769
     }
   ],
   "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": 0.14314839806985447,
-    "producer_p95_seconds": 0.346929008999723,
-    "gpu_update_seconds": 0.9555796714048799,
-    "producer_batches_per_second": 6.985757531928605,
-    "gpu_batches_per_second": 1.0464852172187946,
-    "producer_headroom": 6.675447886874501,
+    "producer_batch_seconds": 0.42587891601573347,
+    "producer_p95_seconds": 0.3640859739971347,
+    "gpu_update_seconds": 0.9644535649536163,
+    "producer_batches_per_second": 2.34808524769293,
+    "gpu_batches_per_second": 1.0368565541546764,
+    "producer_headroom": 2.2646191879524413,
     "sustainable": true,
-    "buffer_seconds": 60.201519298507435,
+    "buffer_seconds": 60.76057459207783,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283063.36612036853,
-    "upload_bytes_per_second": 5021156.664512638,
+    "generation_bytes_per_second": 282212.6957332189,
+    "upload_bytes_per_second": 5102412.195977184,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.73880899151127,
+    "estimated_full_shards_per_hour": 60.55627492902208,
     "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": 1.0722998880009982,
-      "bytes": 987240,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.0394909079986974,
       "bytes": 6071704,
@@ -562,6 +556,12 @@
       "bytes": 1121083,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 0.944058788001712,
+      "bytes": 7283383,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -367,3 +367,4 @@
 {"step": 593800, "loss": 0.6211135387420654, "lr": 0.0003, "grad_norm": 0.16296979909993956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86914619690, "tokens_per_second": 138247.07420084206, "data_wait_s": 1.0039002518169582e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 593900, "loss": 0.5950559377670288, "lr": 0.0003, "grad_norm": 0.17134861608576935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86929188377, "tokens_per_second": 151957.24558113643, "data_wait_s": 1.0534000466577709e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594000, "loss": 0.5627453327178955, "lr": 0.0003, "grad_norm": 0.16188618342335795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86943668856, "tokens_per_second": 150438.6180611216, "data_wait_s": 1.107199932448566e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 594100, "loss": 0.5970128178596497, "lr": 0.0003, "grad_norm": 0.1625259251657857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86958180732, "tokens_per_second": 149836.70428047888, "data_wait_s": 1.5561003237962723e-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 sha256:47d642583d49e2b478e76de76074a5b55ac0e5e8f544fc8433c3598f9ee1a29a
-size 787188
+oid sha256:4561c554664f8262f3c7a7abff6a63dea904394cf324fbb92659aaff010fc560
+size 789306