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": 76267.97352211941,
-      "required_pages_per_second_upper_estimate": 0.00527639537433996
+      "required_token_positions_per_second": 76910.86358896481,
+      "required_pages_per_second_upper_estimate": 0.0053208720008747976
     },
     {
       "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": 9533.496690264927,
-      "required_pages_per_second_upper_estimate": 0.003991475982164657
+      "required_token_positions_per_second": 9613.857948620602,
+      "required_pages_per_second_upper_estimate": 0.0040251215629042774
     },
     {
       "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": 19066.993380529853,
-      "required_pages_per_second_upper_estimate": 0.003971151890735042
+      "required_token_positions_per_second": 19227.715897241203,
+      "required_pages_per_second_upper_estimate": 0.004004626152428221
     },
     {
       "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": 28600.49007079478,
-      "required_pages_per_second_upper_estimate": 0.0017145557093244084
+      "required_token_positions_per_second": 28841.573845861807,
+      "required_pages_per_second_upper_estimate": 0.0017290083135260664
     },
     {
       "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": 28600.49007079478,
-      "required_pages_per_second_upper_estimate": 0.002223562302743454
+      "required_token_positions_per_second": 28841.573845861807,
+      "required_pages_per_second_upper_estimate": 0.002242305505839457
     },
     {
       "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": 28600.49007079478,
-      "required_pages_per_second_upper_estimate": 0.3668563778144814
+      "required_token_positions_per_second": 28841.573845861807,
+      "required_pages_per_second_upper_estimate": 0.36994874162545127
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08480887981255592,
-    "producer_p95_seconds": 0.23899803899985272,
-    "gpu_update_seconds": 0.9666967220338853,
-    "producer_batches_per_second": 11.791218115487364,
-    "gpu_batches_per_second": 1.0344505957318715,
-    "producer_headroom": 11.398531901028203,
+    "producer_batch_seconds": 0.364867139203227,
+    "producer_p95_seconds": 0.22015066399944772,
+    "gpu_update_seconds": 0.958616202699595,
+    "producer_batches_per_second": 2.740723656791167,
+    "gpu_batches_per_second": 1.0431703503277563,
+    "producer_headroom": 2.6273021045220966,
     "sustainable": true,
-    "buffer_seconds": 60.901893488134775,
+    "buffer_seconds": 60.392820770074486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 544929.4097775734,
+    "generation_bytes_per_second": 473828.7723242232,
     "upload_bytes_per_second": 5536036.134034711,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.92916603083994,
+    "estimated_full_shards_per_hour": 101.67262437148115,
     "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": [

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 411900, "loss": 0.5704567432403564, "lr": 0.0003, "grad_norm": 0.16515601156868315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60291764012, "tokens_per_second": 151566.4214560717, "data_wait_s": 8.141001671901904e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 412000, "loss": 0.5596719980239868, "lr": 0.0003, "grad_norm": 0.1739813141639292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60306317165, "tokens_per_second": 150413.91060559504, "data_wait_s": 1.1639000149443746e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 412100, "loss": 0.6097952127456665, "lr": 0.0003, "grad_norm": 0.14925702554629555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60320879151, "tokens_per_second": 150830.78593537412, "data_wait_s": 1.4716000805492513e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 412200, "loss": 0.595787525177002, "lr": 0.0003, "grad_norm": 0.16429479578764294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60335480850, "tokens_per_second": 151121.78452627858, "data_wait_s": 1.2638000043807551e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c132b2e30fed6ae8448697d4a7ebc58fecb4f45b8eb30e6fb8ca43a75c5278ee
-size 320214
+oid sha256:576596c599488336e8349a3d4e8f8eabc8344dc6cfac2ca01426912819114a2b
+size 322332