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": 77257.37336619939,
-      "required_pages_per_second_upper_estimate": 0.00534484435127734
+      "required_token_positions_per_second": 76837.28047546187,
+      "required_pages_per_second_upper_estimate": 0.005315781350346317
     },
     {
       "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": 9657.171670774924,
-      "required_pages_per_second_upper_estimate": 0.00404325611387692
+      "required_token_positions_per_second": 9604.660059432734,
+      "required_pages_per_second_upper_estimate": 0.004021270598775085
     },
     {
       "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": 19314.343341549848,
-      "required_pages_per_second_upper_estimate": 0.004022668364558378
+      "required_token_positions_per_second": 19209.320118865468,
+      "required_pages_per_second_upper_estimate": 0.004000794796921855
     },
     {
       "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": 28971.51501232477,
-      "required_pages_per_second_upper_estimate": 0.0017367980880468481
+      "required_token_positions_per_second": 28813.9801782982,
+      "required_pages_per_second_upper_estimate": 0.001727354115288719
     },
     {
       "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": 28971.51501232477,
-      "required_pages_per_second_upper_estimate": 0.0022524078599811645
+      "required_token_positions_per_second": 28813.9801782982,
+      "required_pages_per_second_upper_estimate": 0.002240160219558103
     },
     {
       "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": 28971.51501232477,
-      "required_pages_per_second_upper_estimate": 0.3716154873888838
+      "required_token_positions_per_second": 28813.9801782982,
+      "required_pages_per_second_upper_estimate": 0.36959479968571723
     }
   ],
   "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": 0.06866931921885566,
-    "producer_p95_seconds": 0.1913880619995325,
-    "gpu_update_seconds": 0.9543166792705962,
-    "producer_batches_per_second": 14.562544253757704,
-    "gpu_batches_per_second": 1.0478701899712373,
-    "producer_headroom": 13.897278873977154,
+    "producer_batch_seconds": 0.08928619083609135,
+    "producer_p95_seconds": 0.2012268389989913,
+    "gpu_update_seconds": 0.9595342201569091,
+    "producer_batches_per_second": 11.199940221839759,
+    "gpu_batches_per_second": 1.042172315476642,
+    "producer_headroom": 10.746725906567013,
     "sustainable": true,
-    "buffer_seconds": 60.12195079404756,
+    "buffer_seconds": 60.450655869885274,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282558.96777140384,
-    "upload_bytes_per_second": 4846346.667821038,
+    "generation_bytes_per_second": 339923.7721260474,
+    "upload_bytes_per_second": 4809647.447631449,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.63057684761607,
+    "estimated_full_shards_per_hour": 72.93972847782199,
     "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": 1.8011739470002794,
-      "bytes": 11923984,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.9351237119999496,
       "bytes": 1136147,
@@ -562,6 +556,12 @@
       "bytes": 11262011,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.2981046400000196,
+      "bytes": 6199716,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -117,3 +117,4 @@
 {"step": 854200, "loss": 0.5001006722450256, "lr": 3e-05, "grad_norm": 0.30517504769892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125030910501, "tokens_per_second": 150413.6043681169, "data_wait_s": 1.2000000424450263e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 854300, "loss": 0.5145289301872253, "lr": 3e-05, "grad_norm": 0.3026922033517351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125045550987, "tokens_per_second": 153166.03047806947, "data_wait_s": 9.100000170292333e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 854400, "loss": 0.5267497301101685, "lr": 3e-05, "grad_norm": 0.28301786113843835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125060224323, "tokens_per_second": 153080.44934251547, "data_wait_s": 1.3670000043930486e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 854500, "loss": 0.5277312397956848, "lr": 3e-05, "grad_norm": 0.30865370008704085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125074912455, "tokens_per_second": 153124.75779856162, "data_wait_s": 1.2423999578459188e-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 sha256:09742fb965816008e66a26ca58654f924913fa3a7bc65e9dab32f51986367a37
-size 253773
+oid sha256:29ac80eb2821f7cc51adaf2aa6ef650043fb0bf2621c088142dc058425068507
+size 255891