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": 77116.64733517324,
-      "required_pages_per_second_upper_estimate": 0.005335108597921572
+      "required_token_positions_per_second": 76883.09666291905,
+      "required_pages_per_second_upper_estimate": 0.005318951020502807
     },
     {
       "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": 9639.580916896655,
-      "required_pages_per_second_upper_estimate": 0.004035891232564801
+      "required_token_positions_per_second": 9610.387082864881,
+      "required_pages_per_second_upper_estimate": 0.004023668383892276
     },
     {
       "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": 19279.16183379331,
-      "required_pages_per_second_upper_estimate": 0.004015340984291443
+      "required_token_positions_per_second": 19220.774165729763,
+      "required_pages_per_second_upper_estimate": 0.004003180372820159
     },
     {
       "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": 28918.742750689966,
-      "required_pages_per_second_upper_estimate": 0.0017336344715404195
+      "required_token_positions_per_second": 28831.161248594646,
+      "required_pages_per_second_upper_estimate": 0.0017283840942190157
     },
     {
       "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": 28918.742750689966,
-      "required_pages_per_second_upper_estimate": 0.0022483050487597063
+      "required_token_positions_per_second": 28831.161248594646,
+      "required_pages_per_second_upper_estimate": 0.0022414959722021105
     },
     {
       "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": 28918.742750689966,
-      "required_pages_per_second_upper_estimate": 0.37093858147907244
+      "required_token_positions_per_second": 28831.161248594646,
+      "required_pages_per_second_upper_estimate": 0.36981518000788405
     }
   ],
   "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": 0.09040783064062907,
-    "producer_p95_seconds": 0.26354931599780684,
-    "gpu_update_seconds": 0.9560581605623347,
-    "producer_batches_per_second": 11.060988776237734,
-    "gpu_batches_per_second": 1.045961471017432,
-    "producer_headroom": 10.574948583410476,
+    "producer_batch_seconds": 0.4955320655235482,
+    "producer_p95_seconds": 0.26281252799890353,
+    "gpu_update_seconds": 0.9589624143685049,
+    "producer_batches_per_second": 2.018032877334512,
+    "gpu_batches_per_second": 1.0427937372900262,
+    "producer_headroom": 1.9352176803237247,
     "sustainable": true,
-    "buffer_seconds": 60.23166411542709,
+    "buffer_seconds": 60.41463210521581,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357718.8962601585,
-    "upload_bytes_per_second": 3755698.008028069,
+    "generation_bytes_per_second": 358437.8700976959,
+    "upload_bytes_per_second": 3782999.478219358,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.7581478677136,
+    "estimated_full_shards_per_hour": 76.91242291639479,
     "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": 0.9403650779986492,
-      "bytes": 1055784,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.317758003999188,
       "bytes": 10010526,
@@ -562,6 +556,12 @@
       "bytes": 5962126,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0608517170003324,
+      "bytes": 4607080,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 667700, "loss": 0.5745958685874939, "lr": 0.0001, "grad_norm": 0.21178127155839305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97734395267, "tokens_per_second": 153873.54285004552, "data_wait_s": 8.281000191345811e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 667800, "loss": 0.5833641886711121, "lr": 0.0001, "grad_norm": 0.19772748002870502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97749062969, "tokens_per_second": 152365.44524410993, "data_wait_s": 1.1034000635845587e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 667900, "loss": 0.6122384071350098, "lr": 0.0001, "grad_norm": 0.19803908538331397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97763835607, "tokens_per_second": 131008.13372610754, "data_wait_s": 1.2407999747665599e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 668000, "loss": 0.48959314823150635, "lr": 0.0001, "grad_norm": 0.18453661882029934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97778281404, "tokens_per_second": 150372.9252406844, "data_wait_s": 6.645997927989811e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99b077d5df561f5e771c18a712b3667efe6bed5afec12a0a089f3cb99a6c32f0
-size 488319
+oid sha256:8e2d156662977486ba1ff3271fb46b4962af979cca102b7d5d984a3ae74fc173
+size 490437