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": 75747.00882799747,
-      "required_pages_per_second_upper_estimate": 0.0052403538279959735
+      "required_token_positions_per_second": 76372.44038302057,
+      "required_pages_per_second_upper_estimate": 0.0052836226341736155
     },
     {
       "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": 9468.376103499684,
-      "required_pages_per_second_upper_estimate": 0.003964211352358539
+      "required_token_positions_per_second": 9546.555047877571,
+      "required_pages_per_second_upper_estimate": 0.003996943243807556
     },
     {
       "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": 18936.75220699937,
-      "required_pages_per_second_upper_estimate": 0.003944026088979361
+      "required_token_positions_per_second": 19093.110095755143,
+      "required_pages_per_second_upper_estimate": 0.003976591313772373
     },
     {
       "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": 28405.128310499054,
-      "required_pages_per_second_upper_estimate": 0.0017028440700948154
+      "required_token_positions_per_second": 28639.665143632716,
+      "required_pages_per_second_upper_estimate": 0.0017169041951241699
     },
     {
       "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": 28405.128310499054,
-      "required_pages_per_second_upper_estimate": 0.002208373785186031
+      "required_token_positions_per_second": 28639.665143632716,
+      "required_pages_per_second_upper_estimate": 0.002226607992343669
     },
     {
       "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": 28405.128310499054,
-      "required_pages_per_second_upper_estimate": 0.36435048691652305
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28639.665143632716,
+      "required_pages_per_second_upper_estimate": 0.36735887358593033
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0685815586561489,
-    "producer_p95_seconds": 0.25341320499865105,
-    "gpu_update_seconds": 0.9733453655895226,
-    "producer_batches_per_second": 14.581179249858618,
-    "gpu_batches_per_second": 1.0273845598415456,
-    "producer_headroom": 14.192523247679997,
+    "producer_batch_seconds": 0.13385324279670385,
+    "producer_p95_seconds": 0.2571261470002355,
+    "gpu_update_seconds": 0.9653744155645902,
+    "producer_batches_per_second": 7.470868685032897,
+    "gpu_batches_per_second": 1.0358675182158823,
+    "producer_headroom": 7.2121854905734315,
     "sustainable": true,
-    "buffer_seconds": 60.3474126665504,
+    "buffer_seconds": 60.818588180569186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276727.3604379929,
-    "upload_bytes_per_second": 5033871.369976651,
+    "generation_bytes_per_second": 288265.61806730065,
+    "upload_bytes_per_second": 4994296.893667651,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.37924966673699,
+    "estimated_full_shards_per_hour": 61.85509115709557,
     "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": 1.4825698250006099,
-      "bytes": 11644292,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.3249330049984565,
       "bytes": 7496918,
@@ -562,6 +556,12 @@
       "bytes": 10168781,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.3043567909990088,
+      "bytes": 7359704,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -181,3 +181,4 @@
 {"step": 738100, "loss": 0.535503625869751, "lr": 5e-05, "grad_norm": 0.2167565098519966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108037535724, "tokens_per_second": 150431.75698752122, "data_wait_s": 1.7815997125580907e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738200, "loss": 0.5587160587310791, "lr": 5e-05, "grad_norm": 0.2236781793592614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108052271803, "tokens_per_second": 132744.06884565984, "data_wait_s": 1.8234000890515745e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738300, "loss": 0.5340388417243958, "lr": 5e-05, "grad_norm": 0.2364735244449206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108067046735, "tokens_per_second": 152113.6151064188, "data_wait_s": 2.4574001145083457e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 738400, "loss": 0.5231951475143433, "lr": 5e-05, "grad_norm": 0.24424534864142355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108081583373, "tokens_per_second": 148619.50765769536, "data_wait_s": 1.7316000594291836e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77eed00476898fc3f8b20ac124bd163087a24655094ccafa79e38a0a3af3bcb6
-size 390108
+oid sha256:cd27927cf22306f691f829b69439f83262c46896db4f3595417cb8b95c16a239
+size 392226