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": 76883.09666291905,
-      "required_pages_per_second_upper_estimate": 0.005318951020502807
+      "required_token_positions_per_second": 76182.42988643878,
+      "required_pages_per_second_upper_estimate": 0.005270477267134993
     },
     {
       "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": 9610.387082864881,
-      "required_pages_per_second_upper_estimate": 0.004023668383892276
+      "required_token_positions_per_second": 9522.803735804848,
+      "required_pages_per_second_upper_estimate": 0.003986999065426503
     },
     {
       "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": 19220.774165729763,
-      "required_pages_per_second_upper_estimate": 0.004003180372820159
+      "required_token_positions_per_second": 19045.607471609695,
+      "required_pages_per_second_upper_estimate": 0.003966697769891317
     },
     {
       "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": 28831.161248594646,
-      "required_pages_per_second_upper_estimate": 0.0017283840942190157
+      "required_token_positions_per_second": 28568.41120741454,
+      "required_pages_per_second_upper_estimate": 0.0017126326304463513
     },
     {
       "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": 28831.161248594646,
-      "required_pages_per_second_upper_estimate": 0.0022414959722021105
+      "required_token_positions_per_second": 28568.41120741454,
+      "required_pages_per_second_upper_estimate": 0.002221068312215649
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28831.161248594646,
-      "required_pages_per_second_upper_estimate": 0.36981518000788405
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28568.41120741454,
+      "required_pages_per_second_upper_estimate": 0.3664449045986396
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "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,
+    "producer_batch_seconds": 0.17731562778149623,
+    "producer_p95_seconds": 0.26181521299804444,
+    "gpu_update_seconds": 0.9677822052919883,
+    "producer_batches_per_second": 5.63966082691982,
+    "gpu_batches_per_second": 1.0332903359163246,
+    "producer_headroom": 5.4579633921753015,
     "sustainable": true,
-    "buffer_seconds": 60.41463210521581,
+    "buffer_seconds": 60.002496728103274,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358437.8700976959,
+    "generation_bytes_per_second": 352146.538909456,
     "upload_bytes_per_second": 3782999.478219358,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.91242291639479,
+    "estimated_full_shards_per_hour": 75.56244969809303,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -228,3 +228,4 @@
 {"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}
+{"step": 668100, "loss": 0.5336928963661194, "lr": 0.0001, "grad_norm": 0.19112198194547503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97792781550, "tokens_per_second": 151095.12986784498, "data_wait_s": 1.194099968415685e-05, "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 sha256:8e2d156662977486ba1ff3271fb46b4962af979cca102b7d5d984a3ae74fc173
-size 490437
+oid sha256:74c78bf4ccf1049229e12943b1430f6c3f7be8d16692695da105ca5a25447c30
+size 492555