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": 77170.4645409262,
-      "required_pages_per_second_upper_estimate": 0.005338831797088684
+      "required_token_positions_per_second": 77144.2485765202,
+      "required_pages_per_second_upper_estimate": 0.005337018115841658
     },
     {
       "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": 9646.308067615775,
-      "required_pages_per_second_upper_estimate": 0.004038707750092015
+      "required_token_positions_per_second": 9643.031072065025,
+      "required_pages_per_second_upper_estimate": 0.0040373357404863645
     },
     {
       "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": 19292.61613523155,
-      "required_pages_per_second_upper_estimate": 0.004018143160467239
+      "required_token_positions_per_second": 19286.06214413005,
+      "required_pages_per_second_upper_estimate": 0.004016778136961165
     },
     {
       "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": 28938.924202847327,
-      "required_pages_per_second_upper_estimate": 0.0017348443187819602
+      "required_token_positions_per_second": 28929.093216195073,
+      "required_pages_per_second_upper_estimate": 0.001734254966143205
     },
     {
       "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": 28938.924202847327,
-      "required_pages_per_second_upper_estimate": 0.0022498740678961143
+      "required_token_positions_per_second": 28929.093216195073,
+      "required_pages_per_second_upper_estimate": 0.002249109751925842
     },
     {
       "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": 28938.924202847327,
-      "required_pages_per_second_upper_estimate": 0.371197447478192
+      "required_token_positions_per_second": 28929.093216195073,
+      "required_pages_per_second_upper_estimate": 0.37107134613710796
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11864098548431912,
-    "producer_p95_seconds": 0.3644821529960609,
-    "gpu_update_seconds": 0.9553914238898932,
-    "producer_batches_per_second": 8.428790404241635,
-    "gpu_batches_per_second": 1.0466914135867813,
-    "producer_headroom": 8.052794065977883,
+    "producer_batch_seconds": 0.1273454619296217,
+    "producer_p95_seconds": 0.3612714100017911,
+    "gpu_update_seconds": 0.9557160949836515,
+    "producer_batches_per_second": 7.852655170018202,
+    "gpu_batches_per_second": 1.0463358368126112,
+    "producer_headroom": 7.504908934342978,
     "sustainable": true,
-    "buffer_seconds": 60.18965970506327,
+    "buffer_seconds": 60.21011398397005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311823.2564891013,
+    "generation_bytes_per_second": 312945.43193868274,
     "upload_bytes_per_second": 5295058.526138552,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.91001196865824,
+    "estimated_full_shards_per_hour": 67.15080469723092,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -405,3 +405,4 @@
 {"step": 480700, "loss": 0.5943543314933777, "lr": 0.0003, "grad_norm": 0.1558133987893216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70358842762, "tokens_per_second": 152542.6546573282, "data_wait_s": 9.530005627311766e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 480800, "loss": 0.5489296913146973, "lr": 0.0003, "grad_norm": 0.17829685378553195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70373700418, "tokens_per_second": 153900.578214714, "data_wait_s": 2.7836998924613e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 480900, "loss": 0.5871713757514954, "lr": 0.0003, "grad_norm": 0.1544097962928106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70388401971, "tokens_per_second": 152389.88821499544, "data_wait_s": 2.2731001081410795e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 481000, "loss": 0.6311573386192322, "lr": 0.0003, "grad_norm": 0.19809674490948792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70402954995, "tokens_per_second": 150483.00139091918, "data_wait_s": 1.1340001947246492e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1580580e613595bf011b545cec7b5cb0852c16856cf602a96913b642351068e
-size 868455
+oid sha256:b97f1c4474f30dea9e55bf40cf93a566b40a4c4ed3b707725276808ec141ba03
+size 870573