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": 76473.22158198687,
-      "required_pages_per_second_upper_estimate": 0.005290594911362703
+      "required_token_positions_per_second": 76758.82913070804,
+      "required_pages_per_second_upper_estimate": 0.005310353904284048
     },
     {
       "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": 9559.152697748359,
-      "required_pages_per_second_upper_estimate": 0.004002217616739611
+      "required_token_positions_per_second": 9594.853641338505,
+      "required_pages_per_second_upper_estimate": 0.004017164856300327
     },
     {
       "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": 19118.305395496718,
-      "required_pages_per_second_upper_estimate": 0.003981838830263807
+      "required_token_positions_per_second": 19189.70728267701,
+      "required_pages_per_second_upper_estimate": 0.003996709960369071
     },
     {
       "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": 28677.458093245077,
-      "required_pages_per_second_upper_estimate": 0.001719169824746938
+      "required_token_positions_per_second": 28784.560924015514,
+      "required_pages_per_second_upper_estimate": 0.0017255904759150723
     },
     {
       "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": 28677.458093245077,
-      "required_pages_per_second_upper_estimate": 0.0022295462279424126
+      "required_token_positions_per_second": 28784.560924015514,
+      "required_pages_per_second_upper_estimate": 0.0022378730019392483
     },
     {
       "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": 28677.458093245077,
-      "required_pages_per_second_upper_estimate": 0.36784364096465
+      "required_token_positions_per_second": 28784.560924015514,
+      "required_pages_per_second_upper_estimate": 0.3692174410797131
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2973658871562179,
-    "producer_p95_seconds": 0.36780067200015765,
-    "gpu_update_seconds": 0.964102184723005,
-    "producer_batches_per_second": 3.3628605135687972,
-    "gpu_batches_per_second": 1.0372344507105424,
-    "producer_headroom": 3.242141168050404,
+    "producer_batch_seconds": 0.3294581948438804,
+    "producer_p95_seconds": 0.3492206829978386,
+    "gpu_update_seconds": 0.9605149118996198,
+    "producer_batches_per_second": 3.035286466235474,
+    "gpu_batches_per_second": 1.0411082510132925,
+    "producer_headroom": 2.915437912706275,
     "sustainable": true,
-    "buffer_seconds": 60.73843763754932,
+    "buffer_seconds": 60.512439449676044,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 480238.167949057,
+    "generation_bytes_per_second": 444893.92860640655,
     "upload_bytes_per_second": 3932823.000799991,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.04793146947654,
+    "estimated_full_shards_per_hour": 95.4638804783263,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -368,3 +368,4 @@
 {"step": 836800, "loss": 0.45729824900627136, "lr": 3e-05, "grad_norm": 0.3163129596705552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122483637692, "tokens_per_second": 152265.28736622035, "data_wait_s": 1.0912001016549766e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 836900, "loss": 0.54319828748703, "lr": 3e-05, "grad_norm": 0.299937191501277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122498248230, "tokens_per_second": 151109.9856754473, "data_wait_s": 1.3634002243634313e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 837000, "loss": 0.52271568775177, "lr": 3e-05, "grad_norm": 0.2812109963595893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122512721369, "tokens_per_second": 149911.48262071869, "data_wait_s": 1.103399699786678e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 837100, "loss": 0.5420165061950684, "lr": 3e-05, "grad_norm": 0.2981183837032969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122527383846, "tokens_per_second": 152028.0753820676, "data_wait_s": 9.708994184620678e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b1502fdd0037a779e1ed601dcb918e041ba7e21c597cbbb93f510ef06760f5a
-size 789306
+oid sha256:58298e67af0a9f6b4fac086d10735d9e1e68ede072030cd324e4a886aaa250f2
+size 791424