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": 76445.07360592528,
-      "required_pages_per_second_upper_estimate": 0.005288647569066459
+      "required_token_positions_per_second": 76614.25414679227,
+      "required_pages_per_second_upper_estimate": 0.005300351871436576
     },
     {
       "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": 9555.63420074066,
-      "required_pages_per_second_upper_estimate": 0.00400074449551706
+      "required_token_positions_per_second": 9576.781768349034,
+      "required_pages_per_second_upper_estimate": 0.004009598540463257
     },
     {
       "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": 19111.26840148132,
-      "required_pages_per_second_upper_estimate": 0.003980373209988412
+      "required_token_positions_per_second": 19153.563536698068,
+      "required_pages_per_second_upper_estimate": 0.003989182171256337
     },
     {
       "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": 28666.90260222198,
-      "required_pages_per_second_upper_estimate": 0.001718537039177405
+      "required_token_positions_per_second": 28730.3453050471,
+      "required_pages_per_second_upper_estimate": 0.0017223403323403728
     },
     {
       "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": 28666.90260222198,
-      "required_pages_per_second_upper_estimate": 0.0022287255849440654
+      "required_token_positions_per_second": 28730.3453050471,
+      "required_pages_per_second_upper_estimate": 0.00223365797603376
     },
     {
       "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": 28666.90260222198,
-      "required_pages_per_second_upper_estimate": 0.3677082464594089
+      "required_token_positions_per_second": 28730.3453050471,
+      "required_pages_per_second_upper_estimate": 0.3685220213317826
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09431194092192641,
-    "producer_p95_seconds": 0.26395059499918716,
-    "gpu_update_seconds": 0.9644571784974424,
-    "producer_batches_per_second": 10.603111230929104,
-    "gpu_batches_per_second": 1.0368526693512001,
-    "producer_headroom": 10.226246741076427,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.76080224533887,
+    "buffer_seconds": 60.62662949247642,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391914.2333870072,
-    "upload_bytes_per_second": 3228721.5237426376,
+    "generation_bytes_per_second": 363477.14959397184,
+    "upload_bytes_per_second": 3205243.6623058766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.09567118842756,
+    "estimated_full_shards_per_hour": 77.99373498787276,
     "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": [
-    {
-      "seconds": 1.5652879460030817,
-      "bytes": 11222900,
-      "pages": 47,
-      "shards": 3
-    },
     {
       "seconds": 1.6341935630007356,
       "bytes": 5648901,
@@ -562,6 +556,12 @@
       "bytes": 6430899,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.6179743659995438,
+      "bytes": 8234542,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -277,3 +277,4 @@
 {"step": 827700, "loss": 0.5382410883903503, "lr": 3e-05, "grad_norm": 0.2850816636258624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121151734190, "tokens_per_second": 149867.66891448153, "data_wait_s": 9.880997822619975e-06, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:321ab665aba02a62c3ecb10aa2ae130a55698b0cb96427c6fed54f12c0fb52ef
-size 595002
+oid sha256:2bbcf10b2849ef9520c0b26409406d24ded3fdb86597edc0fa547770fffeb816
+size 597120