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": 77333.06992617647,
-      "required_pages_per_second_upper_estimate": 0.005350081214936772
+      "required_token_positions_per_second": 77221.22906091632,
+      "required_pages_per_second_upper_estimate": 0.005342343804371514
     },
     {
       "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": 9666.633740772058,
-      "required_pages_per_second_upper_estimate": 0.00404721768499423
+      "required_token_positions_per_second": 9652.65363261454,
+      "required_pages_per_second_upper_estimate": 0.004041364505646533
     },
     {
       "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": 19333.267481544117,
-      "required_pages_per_second_upper_estimate": 0.004026609763856053
+      "required_token_positions_per_second": 19305.30726522908,
+      "required_pages_per_second_upper_estimate": 0.0040207863881581195
     },
     {
       "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": 28999.901222316174,
-      "required_pages_per_second_upper_estimate": 0.0017384997979925999
+      "required_token_positions_per_second": 28957.96089784362,
+      "required_pages_per_second_upper_estimate": 0.0017359855395796377
     },
     {
       "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": 28999.901222316174,
-      "required_pages_per_second_upper_estimate": 0.002254614762950254
+      "required_token_positions_per_second": 28957.96089784362,
+      "required_pages_per_second_upper_estimate": 0.002251354087198504
     },
     {
       "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": 28999.901222316174,
-      "required_pages_per_second_upper_estimate": 0.37197959521191587
+      "required_token_positions_per_second": 28957.96089784362,
+      "required_pages_per_second_upper_estimate": 0.37144162976159384
     }
   ],
   "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": 0.08623498239838057,
-    "producer_p95_seconds": 0.20743652899909648,
-    "gpu_update_seconds": 0.9533825576869258,
-    "producer_batches_per_second": 11.59622199933074,
-    "gpu_batches_per_second": 1.048896890274746,
-    "producer_headroom": 11.055635789227338,
+    "producer_batch_seconds": 0.07832211746087125,
+    "producer_p95_seconds": 0.21099159400000644,
+    "gpu_update_seconds": 0.9547633584262086,
+    "producer_batches_per_second": 12.767785555588272,
+    "gpu_batches_per_second": 1.0473799514555706,
+    "producer_headroom": 12.190213816719094,
     "sustainable": true,
-    "buffer_seconds": 60.06310113427632,
+    "buffer_seconds": 60.15009158085115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444037.14331205323,
-    "upload_bytes_per_second": 5622343.015514161,
+    "generation_bytes_per_second": 443450.12111196015,
+    "upload_bytes_per_second": 5608984.074421353,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.28003429909894,
+    "estimated_full_shards_per_hour": 95.1540730001364,
     "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": 0.7976953110000977,
-      "bytes": 1131427,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1826321499997903,
       "bytes": 12996691,
@@ -562,6 +556,12 @@
       "bytes": 11777262,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.9635194839993346,
+      "bytes": 1029009,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -105,3 +105,4 @@
 {"step": 567600, "loss": 0.6129605174064636, "lr": 0.0003, "grad_norm": 0.15824549312024988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83079676448, "tokens_per_second": 153325.14696044597, "data_wait_s": 1.771099960024003e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 567700, "loss": 0.5564883351325989, "lr": 0.0003, "grad_norm": 0.1546044285232266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83094420347, "tokens_per_second": 153465.20481614032, "data_wait_s": 8.302999049192294e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 567800, "loss": 0.5703026652336121, "lr": 0.0003, "grad_norm": 0.16510065624460302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83109136146, "tokens_per_second": 153065.39365362955, "data_wait_s": 1.9455999790807255e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 567900, "loss": 0.5990357398986816, "lr": 0.0003, "grad_norm": 0.15096866719414792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83123804273, "tokens_per_second": 153176.82087323375, "data_wait_s": 8.633000106783584e-06, "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:ef6169d0a979c18759fe85923c1b054185651e93db3b5fb5fa54e4040d2f5a44
-size 228357
+oid sha256:3f47d964353be417745f6c8984bd1e0530b48074fb76b7204974cd0028f758d6
+size 230475