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": 77300.44987434724,
-      "required_pages_per_second_upper_estimate": 0.005347824484062277
+      "required_token_positions_per_second": 77599.92548299082,
+      "required_pages_per_second_upper_estimate": 0.005368542901547388
     },
     {
       "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": 9662.556234293404,
-      "required_pages_per_second_upper_estimate": 0.004045510518179635
+      "required_token_positions_per_second": 9699.990685373852,
+      "required_pages_per_second_upper_estimate": 0.004061183541126788
     },
     {
       "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": 19325.11246858681,
-      "required_pages_per_second_upper_estimate": 0.004024911289719199
+      "required_token_positions_per_second": 19399.981370747704,
+      "required_pages_per_second_upper_estimate": 0.0040405045076135895
     },
     {
       "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": 28987.668702880215,
-      "required_pages_per_second_upper_estimate": 0.0017377664771303898
+      "required_token_positions_per_second": 29099.972056121554,
+      "required_pages_per_second_upper_estimate": 0.0017444988916799165
     },
     {
       "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": 28987.668702880215,
-      "required_pages_per_second_upper_estimate": 0.002253663738369275
+      "required_token_positions_per_second": 29099.972056121554,
+      "required_pages_per_second_upper_estimate": 0.002262394830113542
     },
     {
       "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": 28987.668702880215,
-      "required_pages_per_second_upper_estimate": 0.37182268958684744
+      "required_token_positions_per_second": 29099.972056121554,
+      "required_pages_per_second_upper_estimate": 0.37326319642028133
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07263681504696251,
-    "producer_p95_seconds": 0.25400387000263436,
-    "gpu_update_seconds": 0.953784876023952,
-    "producer_batches_per_second": 13.767123453216682,
-    "gpu_batches_per_second": 1.0484544525057948,
-    "producer_headroom": 13.130874136032716,
+    "producer_batch_seconds": 0.08993918450008209,
+    "producer_p95_seconds": 0.2573872399989341,
+    "gpu_update_seconds": 0.9501040051405784,
+    "producer_batches_per_second": 11.118624274373838,
+    "gpu_batches_per_second": 1.052516350409489,
+    "producer_headroom": 10.563849454735841,
     "sustainable": true,
-    "buffer_seconds": 60.088447189508976,
+    "buffer_seconds": 60.806656328997015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 512651.1745965123,
+    "generation_bytes_per_second": 528384.8099029668,
     "upload_bytes_per_second": 3819351.2749871835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.00300815984274,
+    "estimated_full_shards_per_hour": 113.3790800363231,
     "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-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 549600, "loss": 0.5847856402397156, "lr": 0.0003, "grad_norm": 0.1970315840782169, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80443946447, "tokens_per_second": 153066.19631061135, "data_wait_s": 1.0759002179838717e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549700, "loss": 0.5437107682228088, "lr": 0.0003, "grad_norm": 0.14214246667607797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80458549955, "tokens_per_second": 152045.07784303802, "data_wait_s": 1.4467997971223667e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549800, "loss": 0.5712734460830688, "lr": 0.0003, "grad_norm": 0.17033618191826674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80473228994, "tokens_per_second": 153308.82749478772, "data_wait_s": 1.3146000128472224e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 549900, "loss": 0.6541333794593811, "lr": 0.0003, "grad_norm": 0.16892563216561565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80487929710, "tokens_per_second": 153447.29287662983, "data_wait_s": 9.311999747296795e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0ef2982d0abcff3fcaab6cdf2d8346a94b8ef6249023d1c56f19fd14760badc
-size 528561
+oid sha256:63e8b4236f4d6fc67900192dbc8fcac1c3bd17c314ba205f509f060d5473deba
+size 530679