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": 77388.04438534321,
-      "required_pages_per_second_upper_estimate": 0.005353884475580247
+      "required_token_positions_per_second": 77587.89077767279,
+      "required_pages_per_second_upper_estimate": 0.005367710312709124
     },
     {
       "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": 9673.505548167901,
-      "required_pages_per_second_upper_estimate": 0.004050094767251213
+      "required_token_positions_per_second": 9698.486347209098,
+      "required_pages_per_second_upper_estimate": 0.0040605537061513585
     },
     {
       "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": 19347.011096335802,
-      "required_pages_per_second_upper_estimate": 0.004029472196373665
+      "required_token_positions_per_second": 19396.972694418197,
+      "required_pages_per_second_upper_estimate": 0.004039877879678283
     },
     {
       "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": 29020.5166445037,
-      "required_pages_per_second_upper_estimate": 0.001739735661074815
+      "required_token_positions_per_second": 29095.459041627295,
+      "required_pages_per_second_upper_estimate": 0.0017442283433519597
     },
     {
       "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": 29020.5166445037,
-      "required_pages_per_second_upper_estimate": 0.0022562175213476747
+      "required_token_positions_per_second": 29095.459041627295,
+      "required_pages_per_second_upper_estimate": 0.0022620439630872667
     },
     {
       "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": 29020.5166445037,
-      "required_pages_per_second_upper_estimate": 0.37224402771262166
+      "required_token_positions_per_second": 29095.459041627295,
+      "required_pages_per_second_upper_estimate": 0.3732053083160464
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09408632047654919,
-    "producer_p95_seconds": 0.14838632900000448,
-    "gpu_update_seconds": 0.9527052994501514,
-    "producer_batches_per_second": 10.6285376549426,
-    "gpu_batches_per_second": 1.0496425290980795,
-    "producer_headroom": 10.1258641492693,
+    "producer_batch_seconds": 0.21879824581250062,
+    "producer_p95_seconds": 0.1498092589999942,
+    "gpu_update_seconds": 0.9502513763554514,
+    "producer_batches_per_second": 4.570420554728537,
+    "gpu_batches_per_second": 1.0523531192718205,
+    "producer_headroom": 4.3430484226540385,
     "sustainable": true,
-    "buffer_seconds": 60.020433865359536,
+    "buffer_seconds": 60.81608808674889,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458613.6055507164,
-    "upload_bytes_per_second": 4877637.7683359925,
+    "generation_bytes_per_second": 423792.63003792457,
+    "upload_bytes_per_second": 4671671.723227538,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.40780377284163,
+    "estimated_full_shards_per_hour": 90.93603301862052,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 7812095,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9189778770000885,
+      "bytes": 1227062,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 887500, "loss": 0.4985582232475281, "lr": 3e-05, "grad_norm": 0.30761058333817004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129905366636, "tokens_per_second": 153305.5516986935, "data_wait_s": 1.6431999938504305e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887600, "loss": 0.49373859167099, "lr": 3e-05, "grad_norm": 0.29812396705866234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129920081715, "tokens_per_second": 153026.20680727615, "data_wait_s": 8.690999948157696e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 887700, "loss": 0.5393361449241638, "lr": 3e-05, "grad_norm": 0.3169755863280241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129934846871, "tokens_per_second": 154480.24010580132, "data_wait_s": 1.3175000276532955e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 887800, "loss": 0.5599460601806641, "lr": 3e-05, "grad_norm": 0.32537648868909574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129949566163, "tokens_per_second": 153708.03141012922, "data_wait_s": 2.7463000151328743e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f3fb2b44bd3dce42e53cb3a26512b9a873b24b1b1c2ffa07c34b41ab19708c6
-size 57351
+oid sha256:1246d37cb1f607328c47c63da74e03362f5a456497b331e3edf5eb10def189b4
+size 59469