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": 75315.19681926185,
-      "required_pages_per_second_upper_estimate": 0.005210480071289742
+      "required_token_positions_per_second": 76398.81493601462,
+      "required_pages_per_second_upper_estimate": 0.0052854472869471375
     },
     {
       "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": 9414.399602407731,
-      "required_pages_per_second_upper_estimate": 0.003941612518508854
+      "required_token_positions_per_second": 9549.851867001827,
+      "required_pages_per_second_upper_estimate": 0.003998323553129471
     },
     {
       "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": 18828.799204815463,
-      "required_pages_per_second_upper_estimate": 0.003921542325536567
+      "required_token_positions_per_second": 19099.703734003655,
+      "required_pages_per_second_upper_estimate": 0.003977964594733605
     },
     {
       "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": 28243.198807223198,
-      "required_pages_per_second_upper_estimate": 0.0016931366436254629
+      "required_token_positions_per_second": 28649.555601005486,
+      "required_pages_per_second_upper_estimate": 0.0017174971129418134
     },
     {
       "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": 28243.198807223198,
-      "required_pages_per_second_upper_estimate": 0.0021957844785589447
+      "required_token_positions_per_second": 28649.555601005486,
+      "required_pages_per_second_upper_estimate": 0.0022273769319008773
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28243.198807223198,
-      "required_pages_per_second_upper_estimate": 0.36227342911485483
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28649.555601005486,
+      "required_pages_per_second_upper_estimate": 0.3674857377535625
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13434041346880576,
-    "producer_p95_seconds": 0.29059723899990786,
-    "gpu_update_seconds": 0.978925942090137,
-    "producer_batches_per_second": 7.443776404873154,
-    "gpu_batches_per_second": 1.0215277346362557,
-    "producer_headroom": 7.286905829848785,
+    "producer_batch_seconds": 0.07828589718772605,
+    "producer_p95_seconds": 0.2794535739994899,
+    "gpu_update_seconds": 0.9650411470616203,
+    "producer_batches_per_second": 12.773692784053367,
+    "gpu_batches_per_second": 1.0362252459854415,
+    "producer_headroom": 12.327139136535603,
     "sustainable": true,
-    "buffer_seconds": 60.69340840958849,
+    "buffer_seconds": 60.79759226488208,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341542.1444139503,
+    "generation_bytes_per_second": 341660.35168130754,
     "upload_bytes_per_second": 4890218.649714651,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.28699349702721,
+    "estimated_full_shards_per_hour": 73.31235802487774,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -249,3 +249,4 @@
 {"step": 867400, "loss": 0.5173141360282898, "lr": 3e-05, "grad_norm": 0.29034363173134775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126963374771, "tokens_per_second": 151831.47326916485, "data_wait_s": 8.228002116084099e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 867500, "loss": 0.506275475025177, "lr": 3e-05, "grad_norm": 0.31691368830607447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126978094571, "tokens_per_second": 153349.89834760778, "data_wait_s": 1.3564000255428255e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 867600, "loss": 0.5680419206619263, "lr": 3e-05, "grad_norm": 0.3018265213378402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126992868980, "tokens_per_second": 152763.10637275715, "data_wait_s": 1.3738001143792644e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 867700, "loss": 0.5103414058685303, "lr": 3e-05, "grad_norm": 0.30515081747424244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127007644107, "tokens_per_second": 139813.26474671107, "data_wait_s": 1.3945998944109306e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e26a0a008c1a3f108e0920bf81680fac172493edf6663cac3b0eb377b416950b
-size 535698
+oid sha256:d97267e4738b5b2760292e4df80c8f3a3096e74e1ab2d29b3c977bf83e792f8e
+size 537816