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": 76878.49904012171,
-      "required_pages_per_second_upper_estimate": 0.005318632946289732
+      "required_token_positions_per_second": 77100.7867627799,
+      "required_pages_per_second_upper_estimate": 0.005334011326721803
     },
     {
       "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": 9609.812380015213,
-      "required_pages_per_second_upper_estimate": 0.00402342776781028
+      "required_token_positions_per_second": 9637.598345347487,
+      "required_pages_per_second_upper_estimate": 0.0040350611712579665
     },
     {
       "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": 19219.624760030427,
-      "required_pages_per_second_upper_estimate": 0.004002940981924854
+      "required_token_positions_per_second": 19275.196690694975,
+      "required_pages_per_second_upper_estimate": 0.004014515149551915
     },
     {
       "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": 28829.43714004564,
-      "required_pages_per_second_upper_estimate": 0.001728280736544064
+      "required_token_positions_per_second": 28912.795036042462,
+      "required_pages_per_second_upper_estimate": 0.001733277915128936
     },
     {
       "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": 28829.43714004564,
-      "required_pages_per_second_upper_estimate": 0.002241361930345977
+      "required_token_positions_per_second": 28912.795036042462,
+      "required_pages_per_second_upper_estimate": 0.002247842640107095
     },
     {
       "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": 28829.43714004564,
-      "required_pages_per_second_upper_estimate": 0.3697930649946209
+      "required_token_positions_per_second": 28912.795036042462,
+      "required_pages_per_second_upper_estimate": 0.3708622905817327
     }
   ],
   "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": 0.4422479172734768,
-    "producer_p95_seconds": 0.31501339000169537,
-    "gpu_update_seconds": 0.959019763920241,
-    "producer_batches_per_second": 2.2611751484668297,
-    "gpu_batches_per_second": 1.0427313780398453,
-    "producer_headroom": 2.168511657064975,
+    "producer_batch_seconds": 0.08189532783606523,
+    "producer_p95_seconds": 0.27555089499946916,
+    "gpu_update_seconds": 0.9562548333889623,
+    "producer_batches_per_second": 12.210708796498864,
+    "gpu_batches_per_second": 1.0457463482364895,
+    "producer_headroom": 11.676549305757158,
     "sustainable": true,
-    "buffer_seconds": 60.41824512697519,
+    "buffer_seconds": 60.24405450350463,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344401.48536576756,
-    "upload_bytes_per_second": 5981670.814151797,
+    "generation_bytes_per_second": 344028.5098457548,
+    "upload_bytes_per_second": 5996899.902374868,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.90054150323648,
+    "estimated_full_shards_per_hour": 73.82050963906748,
     "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": 1.249853739001992,
-      "bytes": 7694709,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.2034739760019875,
       "bytes": 5697150,
@@ -562,6 +556,12 @@
       "bytes": 12260097,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 0.9388873019997845,
+      "bytes": 6924644,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -284,3 +284,4 @@
 {"step": 425500, "loss": 0.5917612314224243, "lr": 0.0003, "grad_norm": 0.15438423566163761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62281491420, "tokens_per_second": 150683.27652891845, "data_wait_s": 1.9363000319572166e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 425600, "loss": 0.6000295877456665, "lr": 0.0003, "grad_norm": 0.1545670994982086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62296119078, "tokens_per_second": 151801.25339343923, "data_wait_s": 8.053997589740902e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 425700, "loss": 0.6632568836212158, "lr": 0.0003, "grad_norm": 0.17485261796040918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62310857885, "tokens_per_second": 152475.5499211867, "data_wait_s": 8.650000381749123e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 425800, "loss": 0.60506272315979, "lr": 0.0003, "grad_norm": 0.16878657541124026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62325561358, "tokens_per_second": 151809.75710584078, "data_wait_s": 1.8686998373595998e-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 sha256:06341c25439deaf3f3a54fd318ed2e7c1dbdc747dfb3bf4610af0ae106af30c4
-size 610611
+oid sha256:f682054a74a165cdeedef476aaea1228887a333ee3fa5d3b9c964bccf493b946
+size 612729