Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75414.05539644208,
-      "required_pages_per_second_upper_estimate": 0.00653669918658764
+      "required_token_positions_per_second": 76199.15947904534,
+      "required_pages_per_second_upper_estimate": 0.006604750018639585
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9426.75692455526,
-      "required_pages_per_second_upper_estimate": 0.0037604742797810995
+      "required_token_positions_per_second": 9524.894934880667,
+      "required_pages_per_second_upper_estimate": 0.0037996229993939156
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18853.51384911052,
-      "required_pages_per_second_upper_estimate": 0.004361428618480711
+      "required_token_positions_per_second": 19049.789869761335,
+      "required_pages_per_second_upper_estimate": 0.004406833621518293
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28280.27077366578,
-      "required_pages_per_second_upper_estimate": 0.0017221201150025296
+      "required_token_positions_per_second": 28574.684804642002,
+      "required_pages_per_second_upper_estimate": 0.0017400483848179393
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28280.27077366578,
-      "required_pages_per_second_upper_estimate": 0.002177093528956743
+      "required_token_positions_per_second": 28574.684804642002,
+      "required_pages_per_second_upper_estimate": 0.002199758336051492
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28280.27077366578,
-      "required_pages_per_second_upper_estimate": 0.3511463150311755
+      "required_token_positions_per_second": 28574.684804642002,
+      "required_pages_per_second_upper_estimate": 0.35480195195552355
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12353184063266553,
-    "producer_p95_seconds": 0.3670654960005777,
-    "gpu_update_seconds": 0.9776426902441634,
-    "producer_batches_per_second": 8.095078927655596,
-    "gpu_batches_per_second": 1.0228685899039995,
-    "producer_headroom": 7.914094740572053,
+    "producer_batch_seconds": 0.22949615428123593,
+    "producer_p95_seconds": 0.3709906910007703,
+    "gpu_update_seconds": 0.9675697278560546,
+    "producer_batches_per_second": 4.357371491177802,
+    "gpu_batches_per_second": 1.0335172455382668,
+    "producer_headroom": 4.216060747886637,
     "sustainable": true,
-    "buffer_seconds": 60.61384679513813,
+    "buffer_seconds": 60.95689285493144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 701247.7070841481,
+    "generation_bytes_per_second": 677804.9235334038,
     "upload_bytes_per_second": 5103484.385830127,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.23571686455408,
+    "estimated_full_shards_per_hour": 72.7205790495948,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -299,3 +299,4 @@
 {"step": 144900, "loss": 0.6100160982459784, "lr": 0.0003, "grad_norm": 0.13386735083109225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21206014750, "tokens_per_second": 148411.26772408662, "data_wait_s": 1.2540996976895258e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145000, "loss": 0.6024244278669357, "lr": 0.0003, "grad_norm": 0.14424647653323605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21220526567, "tokens_per_second": 148749.33688451187, "data_wait_s": 1.3396002032095566e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145100, "loss": 0.5617381744086742, "lr": 0.0003, "grad_norm": 0.12929500746264588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21235033207, "tokens_per_second": 139643.53645174686, "data_wait_s": 2.4274999304907396e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 145200, "loss": 0.6110075414180756, "lr": 0.0003, "grad_norm": 0.13060606311302397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21249704512, "tokens_per_second": 149717.61129311027, "data_wait_s": 1.4770001143915579e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3deb22457be2df8d8dd9279814cd93c4a3157a861ed1b003f9914fb3e01a195f
-size 642381
+oid sha256:de5d12b59f820753d3321932f36b737190e2874338272b8d1feae7a2344dcee6
+size 644499