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": 77277.85989363043,
-      "required_pages_per_second_upper_estimate": 0.005346261656780312
+      "required_token_positions_per_second": 76565.8749729113,
+      "required_pages_per_second_upper_estimate": 0.0052970048879323415
     },
     {
       "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": 9659.732486703804,
-      "required_pages_per_second_upper_estimate": 0.004044328274030425
+      "required_token_positions_per_second": 9570.734371613913,
+      "required_pages_per_second_upper_estimate": 0.004007066621734267
     },
     {
       "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": 19319.46497340761,
-      "required_pages_per_second_upper_estimate": 0.004023735065407835
+      "required_token_positions_per_second": 19141.468743227826,
+      "required_pages_per_second_upper_estimate": 0.003986663144737639
     },
     {
       "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": 28979.19746011141,
-      "required_pages_per_second_upper_estimate": 0.0017372586390612386
+      "required_token_positions_per_second": 28712.203114841737,
+      "required_pages_per_second_upper_estimate": 0.0017212527357390818
     },
     {
       "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": 28979.19746011141,
-      "required_pages_per_second_upper_estimate": 0.002253005136505059
+      "required_token_positions_per_second": 28712.203114841737,
+      "required_pages_per_second_upper_estimate": 0.0022322475005443544
     },
     {
       "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": 28979.19746011141,
-      "required_pages_per_second_upper_estimate": 0.37171402958032107
+      "required_token_positions_per_second": 28712.203114841737,
+      "required_pages_per_second_upper_estimate": 0.36828931279539434
     }
   ],
   "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": 0.0879581468512356,
-    "producer_p95_seconds": 0.30950272899644915,
-    "gpu_update_seconds": 0.9540636878594121,
-    "producer_batches_per_second": 11.369043525795387,
-    "gpu_batches_per_second": 1.0481480562829648,
-    "producer_headroom": 10.84679159365452,
+    "producer_batch_seconds": 0.14055566585921042,
+    "producer_p95_seconds": 0.3310400049958844,
+    "gpu_update_seconds": 0.9629355117548735,
+    "producer_batches_per_second": 7.114618922595865,
+    "gpu_batches_per_second": 1.0384911427532457,
+    "producer_headroom": 6.850919213170755,
     "sustainable": true,
-    "buffer_seconds": 60.106012335142964,
+    "buffer_seconds": 60.66493724055703,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379456.0719367943,
+    "generation_bytes_per_second": 393323.22698391194,
     "upload_bytes_per_second": 4849181.114898905,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.42243975236771,
+    "estimated_full_shards_per_hour": 84.39800841463106,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -340,3 +340,4 @@
 {"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}
 {"step": 591200, "loss": 0.6221915483474731, "lr": 0.0003, "grad_norm": 0.1789330402481506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86533825638, "tokens_per_second": 152774.5713125196, "data_wait_s": 2.0605999452527612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 591300, "loss": 0.6259649991989136, "lr": 0.0003, "grad_norm": 0.1636940153074632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86548471983, "tokens_per_second": 152535.70387104695, "data_wait_s": 1.1028001608792692e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 591400, "loss": 0.5447350740432739, "lr": 0.0003, "grad_norm": 0.16897899380884174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86563185633, "tokens_per_second": 152892.044087327, "data_wait_s": 1.0538999049458653e-05, "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 sha256:22e222384897804c3ca209a02c43cd87cd771e65284ade85f1e2e8d17bea5772
-size 730002
+oid sha256:536c5807236df3d003a0d7a61fe5d0615fa6962697a8db0275803b815ab0509f
+size 732120