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": 77310.1524053053,
-      "required_pages_per_second_upper_estimate": 0.00534849572766693
+      "required_token_positions_per_second": 77227.48593901488,
+      "required_pages_per_second_upper_estimate": 0.005342776669716331
     },
     {
       "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": 9663.769050663162,
-      "required_pages_per_second_upper_estimate": 0.004046018299067167
+      "required_token_positions_per_second": 9653.43574237686,
+      "required_pages_per_second_upper_estimate": 0.004041691958671707
     },
     {
       "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": 19327.538101326325,
-      "required_pages_per_second_upper_estimate": 0.004025416485050603
+      "required_token_positions_per_second": 19306.87148475372,
+      "required_pages_per_second_upper_estimate": 0.004021112173833859
     },
     {
       "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": 28991.30715198949,
-      "required_pages_per_second_upper_estimate": 0.0017379845965989009
+      "required_token_positions_per_second": 28960.30722713058,
+      "required_pages_per_second_upper_estimate": 0.0017361261984377536
     },
     {
       "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": 28991.30715198949,
-      "required_pages_per_second_upper_estimate": 0.0022539466117836747
+      "required_token_positions_per_second": 28960.30722713058,
+      "required_pages_per_second_upper_estimate": 0.0022515365039801533
     },
     {
       "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": 28991.30715198949,
-      "required_pages_per_second_upper_estimate": 0.37186935970535895
+      "required_token_positions_per_second": 28960.30722713058,
+      "required_pages_per_second_upper_estimate": 0.37147172595439487
     }
   ],
   "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": 0.06748477558596022,
-    "producer_p95_seconds": 0.16395461600041017,
-    "gpu_update_seconds": 0.9536651747040213,
-    "producer_batches_per_second": 14.818157003815296,
-    "gpu_batches_per_second": 1.0485860515042495,
-    "producer_headroom": 14.131560287835132,
+    "producer_batch_seconds": 0.06804180535158366,
+    "producer_p95_seconds": 0.15982736800015118,
+    "gpu_update_seconds": 0.9546860046463462,
+    "producer_batches_per_second": 14.696846958025713,
+    "gpu_batches_per_second": 1.0474648157961002,
+    "producer_headroom": 14.030874103256375,
     "sustainable": true,
-    "buffer_seconds": 60.080906006353345,
+    "buffer_seconds": 60.14521829271981,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436864.73328538216,
-    "upload_bytes_per_second": 3322824.5165877314,
+    "generation_bytes_per_second": 454959.03344075487,
+    "upload_bytes_per_second": 3374851.0387212974,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.7410020725355,
+    "estimated_full_shards_per_hour": 97.62361767212853,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 2742758,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.046531567000784,
+      "bytes": 5904437,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 529400, "loss": 0.5777540802955627, "lr": 0.0003, "grad_norm": 0.18941419501930984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77486755452, "tokens_per_second": 153461.67449992851, "data_wait_s": 1.189700014947448e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 529500, "loss": 0.5765049457550049, "lr": 0.0003, "grad_norm": 0.1820989624634915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77501506326, "tokens_per_second": 153654.1224225014, "data_wait_s": 1.3766999472863972e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 529600, "loss": 0.5900040864944458, "lr": 0.0003, "grad_norm": 0.17206534298399126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77516048745, "tokens_per_second": 151456.50016157047, "data_wait_s": 1.1076000191678759e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 529700, "loss": 0.5776903629302979, "lr": 0.0003, "grad_norm": 0.16426839846052832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77530597653, "tokens_per_second": 151204.83870092846, "data_wait_s": 1.902499934658408e-05, "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:fed23da69fdff9cc40cb9c9eb8c26b8636f7cf37636d654cc5ae51f58f875b1e
-size 97593
+oid sha256:3542f3224dcc9cdfd6b600e381be388f02c6c468649c45112e33cd491ba956c8
+size 99711