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": 76614.25414679227,
-      "required_pages_per_second_upper_estimate": 0.005300351871436576
+      "required_token_positions_per_second": 77012.60287680007,
+      "required_pages_per_second_upper_estimate": 0.005327910560874912
     },
     {
       "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": 9576.781768349034,
-      "required_pages_per_second_upper_estimate": 0.004009598540463257
+      "required_token_positions_per_second": 9626.575359600009,
+      "required_pages_per_second_upper_estimate": 0.004030446077311615
     },
     {
       "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": 19153.563536698068,
-      "required_pages_per_second_upper_estimate": 0.003989182171256337
+      "required_token_positions_per_second": 19253.150719200017,
+      "required_pages_per_second_upper_estimate": 0.00400992355508088
     },
     {
       "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": 28730.3453050471,
-      "required_pages_per_second_upper_estimate": 0.0017223403323403728
+      "required_token_positions_per_second": 28879.726078800024,
+      "required_pages_per_second_upper_estimate": 0.001731295481635626
     },
     {
       "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": 28730.3453050471,
-      "required_pages_per_second_upper_estimate": 0.00223365797603376
+      "required_token_positions_per_second": 28879.726078800024,
+      "required_pages_per_second_upper_estimate": 0.002245271674136466
     },
     {
       "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": 28730.3453050471,
-      "required_pages_per_second_upper_estimate": 0.3685220213317826
+      "required_token_positions_per_second": 28879.726078800024,
+      "required_pages_per_second_upper_estimate": 0.3704381175049066
     }
   ],
   "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": 0.38766616656252495,
-    "producer_p95_seconds": 0.27193547699789633,
-    "gpu_update_seconds": 0.9623274522615305,
-    "producer_batches_per_second": 2.57953901127638,
-    "gpu_batches_per_second": 1.039147327294817,
-    "producer_headroom": 2.482361204730826,
+    "producer_batch_seconds": 0.0931853278047754,
+    "producer_p95_seconds": 0.2728256529990176,
+    "gpu_update_seconds": 0.9573498004988279,
+    "producer_batches_per_second": 10.731303130628186,
+    "gpu_batches_per_second": 1.0445502777343758,
+    "producer_headroom": 10.273610911199341,
     "sustainable": true,
-    "buffer_seconds": 60.62662949247642,
+    "buffer_seconds": 60.31303743142616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363477.14959397184,
+    "generation_bytes_per_second": 378912.22002604546,
     "upload_bytes_per_second": 3205243.6623058766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.99373498787276,
+    "estimated_full_shards_per_hour": 81.30574179254553,
     "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

@@ -278,3 +278,4 @@
 {"step": 827800, "loss": 0.5772397518157959, "lr": 3e-05, "grad_norm": 0.30076087097337906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121166486746, "tokens_per_second": 152557.78921766672, "data_wait_s": 1.3916000170866027e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 827900, "loss": 0.548532247543335, "lr": 3e-05, "grad_norm": 0.31025993526536544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121181269986, "tokens_per_second": 153483.24217754387, "data_wait_s": 5.312002031132579e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 828000, "loss": 0.5721879005432129, "lr": 3e-05, "grad_norm": 0.27754097871963784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121196082024, "tokens_per_second": 153361.0438010824, "data_wait_s": 1.3011002010898665e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 828100, "loss": 0.5479229092597961, "lr": 3e-05, "grad_norm": 0.2848581981503589, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121210710277, "tokens_per_second": 151588.3865879714, "data_wait_s": 8.810002327663824e-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 sha256:2bbcf10b2849ef9520c0b26409406d24ded3fdb86597edc0fa547770fffeb816
-size 597120
+oid sha256:2313923d4a07181008e09203eec6443f96f15cb7200b84bdfd44eec9486fe6e8
+size 599238