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": 76098.65846053648,
-      "required_pages_per_second_upper_estimate": 0.005264681765514583
+      "required_token_positions_per_second": 77254.27510501699,
+      "required_pages_per_second_upper_estimate": 0.005344630006379785
     },
     {
       "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": 9512.33230756706,
-      "required_pages_per_second_upper_estimate": 0.003982614897091628
+      "required_token_positions_per_second": 9656.784388127124,
+      "required_pages_per_second_upper_estimate": 0.004043093966719667
     },
     {
       "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": 19024.66461513412,
-      "required_pages_per_second_upper_estimate": 0.003962335925187721
+      "required_token_positions_per_second": 19313.568776254247,
+      "required_pages_per_second_upper_estimate": 0.004022507043033965
     },
     {
       "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": 28536.99692270118,
-      "required_pages_per_second_upper_estimate": 0.0017107493920446206
+      "required_token_positions_per_second": 28970.35316438137,
+      "required_pages_per_second_upper_estimate": 0.0017367284370366804
     },
     {
       "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": 28536.99692270118,
-      "required_pages_per_second_upper_estimate": 0.0022186259897560203
+      "required_token_positions_per_second": 28970.35316438137,
+      "required_pages_per_second_upper_estimate": 0.002252317531413994
     },
     {
       "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": 28536.99692270118,
-      "required_pages_per_second_upper_estimate": 0.36604195588436755
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28970.35316438137,
+      "required_pages_per_second_upper_estimate": 0.371600584450961
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 704566,
-    "producer_batch_seconds": 0.11071569065580888,
-    "producer_p95_seconds": 0.3281187750035315,
-    "gpu_update_seconds": 0.9688475656668,
-    "producer_batches_per_second": 9.032143448472752,
-    "gpu_batches_per_second": 1.032154113234273,
-    "producer_headroom": 8.750770192806161,
+    "producer_batch_seconds": 0.34750558124164854,
+    "producer_p95_seconds": 0.3360267919997568,
+    "gpu_update_seconds": 0.9543549518751748,
+    "producer_batches_per_second": 2.8776516233983007,
+    "gpu_batches_per_second": 1.0478281671144882,
+    "producer_headroom": 2.746301076561804,
     "sustainable": true,
-    "buffer_seconds": 60.0685490713416,
+    "buffer_seconds": 60.124361968136014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384852.16732313135,
-    "upload_bytes_per_second": 4899088.933693607,
+    "generation_bytes_per_second": 380821.5924632007,
+    "upload_bytes_per_second": 4908349.428134436,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.58031620760399,
+    "estimated_full_shards_per_hour": 81.71544866964355,
     "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": 1.1432905639994715,
-      "bytes": 5745742,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.232796011998289,
       "bytes": 13189628,
@@ -562,6 +556,12 @@
       "bytes": 1086878,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0723825640016003,
+      "bytes": 6205162,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -337,3 +337,4 @@
 {"step": 590800, "loss": 0.5838284492492676, "lr": 0.0003, "grad_norm": 0.17198594263842423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86475194571, "tokens_per_second": 150726.01126480653, "data_wait_s": 1.0618001397233456e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590900, "loss": 0.5800275802612305, "lr": 0.0003, "grad_norm": 0.18188725123569327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86489847229, "tokens_per_second": 152318.04262182084, "data_wait_s": 7.337002898566425e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 591000, "loss": 0.5615832805633545, "lr": 0.0003, "grad_norm": 0.19398214386955098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86504490178, "tokens_per_second": 152016.46038496826, "data_wait_s": 1.4429002476390451e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 591100, "loss": 0.5808337330818176, "lr": 0.0003, "grad_norm": 0.1530831671965875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86519157049, "tokens_per_second": 152745.57757244832, "data_wait_s": 9.272996976505965e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ecec109f8976e1178e202149f125494f4da229a3ba9c59e4a551689816dbc2f9
-size 723648
+oid sha256:9fc7e29470d29905de5cc41b98ef4ae30abe6282b804c4ad057a77b814342b0c
+size 725766