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": 77059.50906093052,
-      "required_pages_per_second_upper_estimate": 0.005331155639530399
+      "required_token_positions_per_second": 77275.73364019739,
+      "required_pages_per_second_upper_estimate": 0.005346114557634227
     },
     {
       "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": 9632.438632616315,
-      "required_pages_per_second_upper_estimate": 0.004032900907284478
+      "required_token_positions_per_second": 9659.466705024674,
+      "required_pages_per_second_upper_estimate": 0.004044216996791525
     },
     {
       "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": 19264.87726523263,
-      "required_pages_per_second_upper_estimate": 0.004012365885369658
+      "required_token_positions_per_second": 19318.933410049347,
+      "required_pages_per_second_upper_estimate": 0.004023624354778577
     },
     {
       "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": 28897.315897848945,
-      "required_pages_per_second_upper_estimate": 0.0017323499644295102
+      "required_token_positions_per_second": 28978.40011507402,
+      "required_pages_per_second_upper_estimate": 0.0017372108394437225
     },
     {
       "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": 28897.315897848945,
-      "required_pages_per_second_upper_estimate": 0.002246639205198082
+      "required_token_positions_per_second": 28978.40011507402,
+      "required_pages_per_second_upper_estimate": 0.0022529431464355524
     },
     {
       "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": 28897.315897848945,
-      "required_pages_per_second_upper_estimate": 0.37066374081718995
+      "required_token_positions_per_second": 28978.40011507402,
+      "required_pages_per_second_upper_estimate": 0.3717038020943038
     }
   ],
   "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": 0.07920860003123664,
-    "producer_p95_seconds": 0.19617519500025082,
-    "gpu_update_seconds": 0.956767060917864,
-    "producer_batches_per_second": 12.624891736574574,
-    "gpu_batches_per_second": 1.045186483573819,
-    "producer_headroom": 12.079080561208682,
+    "producer_batch_seconds": 0.11381758155457078,
+    "producer_p95_seconds": 0.20741610899858642,
+    "gpu_update_seconds": 0.9540899390652705,
+    "producer_batches_per_second": 8.785988828277306,
+    "gpu_batches_per_second": 1.048119217125073,
+    "producer_headroom": 8.382623545799243,
     "sustainable": true,
-    "buffer_seconds": 60.276324837825435,
+    "buffer_seconds": 60.10766616111204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 234047.64120927642,
-    "upload_bytes_per_second": 5156787.556580235,
+    "generation_bytes_per_second": 244511.1267842055,
+    "upload_bytes_per_second": 5212919.988957898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.22117545326919,
+    "estimated_full_shards_per_hour": 52.46639588017105,
     "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": 0.9498594410006262,
-      "bytes": 1178438,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.210409072999937,
       "bytes": 6096975,
@@ -562,6 +556,12 @@
       "bytes": 6026166,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.3405283600004623,
+      "bytes": 7840338,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 451500, "loss": 0.6366727352142334, "lr": 0.0003, "grad_norm": 0.1784375598853069, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66085495661, "tokens_per_second": 129471.76160053082, "data_wait_s": 1.1808999261120334e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 451600, "loss": 0.5924786925315857, "lr": 0.0003, "grad_norm": 0.17195255135016568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66100172617, "tokens_per_second": 153159.60924996025, "data_wait_s": 1.4142000509309582e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 451700, "loss": 0.5548248887062073, "lr": 0.0003, "grad_norm": 0.1487185443346939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66114813112, "tokens_per_second": 152545.10291194072, "data_wait_s": 1.598900053068064e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 451800, "loss": 0.5840235948562622, "lr": 0.0003, "grad_norm": 0.15363027850533936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66129463393, "tokens_per_second": 152427.4422108863, "data_wait_s": 8.916998922359198e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03358bad6adc3f1d45e35169488c2c6de55d392336ada54bd6a54b3acaa06cf3
-size 245301
+oid sha256:7c1ccae3debe1bbb9d81e0697505da3fb029d376269ebe521d3550ca17420be2
+size 247419