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": 76415.03126603276,
-      "required_pages_per_second_upper_estimate": 0.005286569170284862
+      "required_token_positions_per_second": 76453.55884038845,
+      "required_pages_per_second_upper_estimate": 0.005289234597275078
     },
     {
       "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": 9551.878908254095,
-      "required_pages_per_second_upper_estimate": 0.00399917223297236
+      "required_token_positions_per_second": 9556.694855048556,
+      "required_pages_per_second_upper_estimate": 0.004001188569326796
     },
     {
       "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": 19103.75781650819,
-      "required_pages_per_second_upper_estimate": 0.003978808953205954
+      "required_token_positions_per_second": 19113.38971009711,
+      "required_pages_per_second_upper_estimate": 0.003980815022630411
     },
     {
       "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": 28655.636724762284,
-      "required_pages_per_second_upper_estimate": 0.001717861667025695
+      "required_token_positions_per_second": 28670.08456514567,
+      "required_pages_per_second_upper_estimate": 0.001718727792996104
     },
     {
       "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": 28655.636724762284,
-      "required_pages_per_second_upper_estimate": 0.002227849712524932
+      "required_token_positions_per_second": 28670.08456514567,
+      "required_pages_per_second_upper_estimate": 0.0022289729685654063
     },
     {
       "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": 28655.636724762284,
-      "required_pages_per_second_upper_estimate": 0.36756373987971275
+      "required_token_positions_per_second": 28670.08456514567,
+      "required_pages_per_second_upper_estimate": 0.3677490612632684
     }
   ],
   "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": 0.059184239976673325,
-    "producer_p95_seconds": 0.19385498600058781,
-    "gpu_update_seconds": 0.9648363519386902,
-    "producer_batches_per_second": 16.89638999156087,
-    "gpu_batches_per_second": 1.0364451940379877,
-    "producer_headroom": 16.302251280390987,
+    "producer_batch_seconds": 0.07151614278897966,
+    "producer_p95_seconds": 0.19216375399992103,
+    "gpu_update_seconds": 0.964350137760381,
+    "producer_batches_per_second": 13.982857030624082,
+    "gpu_batches_per_second": 1.0369677577092618,
+    "producer_headroom": 13.484370103766045,
     "sustainable": true,
-    "buffer_seconds": 60.78469017213748,
+    "buffer_seconds": 60.754058678904,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315552.6138915958,
+    "generation_bytes_per_second": 357590.7798999976,
     "upload_bytes_per_second": 4034032.713175629,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.71024525223642,
+    "estimated_full_shards_per_hour": 76.7306570792193,
     "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

@@ -78,3 +78,4 @@
 {"step": 807800, "loss": 0.44607335329055786, "lr": 3e-05, "grad_norm": 0.316247497676291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118238535164, "tokens_per_second": 150838.82254082512, "data_wait_s": 1.5902000086498447e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 807900, "loss": 0.5320748686790466, "lr": 3e-05, "grad_norm": 0.2665150959218989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118253232255, "tokens_per_second": 151676.91882096342, "data_wait_s": 1.7311998817604035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808000, "loss": 0.5557799339294434, "lr": 3e-05, "grad_norm": 0.2745567288467813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118267921434, "tokens_per_second": 151798.22802206627, "data_wait_s": 1.627300116524566e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 808100, "loss": 0.5016139149665833, "lr": 3e-05, "grad_norm": 0.27525077283135146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118282451053, "tokens_per_second": 149272.0794929306, "data_wait_s": 1.569300002302043e-05, "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 sha256:7ac8eff8ef3d7bc0fb78eff31835b321eda25e610876cedc8becb4f25a59229c
-size 170388
+oid sha256:9d5df0f318e25a2bda0ba5c8bf4133cf8e4ccbd2cde951310d337db7c1218dc6
+size 172506