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": 76453.55884038845,
-      "required_pages_per_second_upper_estimate": 0.005289234597275078
+      "required_token_positions_per_second": 76159.7479046999,
+      "required_pages_per_second_upper_estimate": 0.005268908075008849
     },
     {
       "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": 9556.694855048556,
-      "required_pages_per_second_upper_estimate": 0.004001188569326796
+      "required_token_positions_per_second": 9519.968488087488,
+      "required_pages_per_second_upper_estimate": 0.00398581200641395
     },
     {
       "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": 19113.38971009711,
-      "required_pages_per_second_upper_estimate": 0.003980815022630411
+      "required_token_positions_per_second": 19039.936976174977,
+      "required_pages_per_second_upper_estimate": 0.003965516755233287
     },
     {
       "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": 28670.08456514567,
-      "required_pages_per_second_upper_estimate": 0.001718727792996104
+      "required_token_positions_per_second": 28559.90546426246,
+      "required_pages_per_second_upper_estimate": 0.0017121227241319017
     },
     {
       "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": 28670.08456514567,
-      "required_pages_per_second_upper_estimate": 0.0022289729685654063
+      "required_token_positions_per_second": 28559.90546426246,
+      "required_pages_per_second_upper_estimate": 0.0022204070280983845
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28670.08456514567,
-      "required_pages_per_second_upper_estimate": 0.3677490612632684
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28559.90546426246,
+      "required_pages_per_second_upper_estimate": 0.3663358020582402
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "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,
+    "producer_batch_seconds": 0.36426121431249214,
+    "producer_p95_seconds": 0.19511630300075922,
+    "gpu_update_seconds": 0.9680704312763379,
+    "producer_batches_per_second": 2.7452826727308954,
+    "gpu_batches_per_second": 1.0329826918497709,
+    "producer_headroom": 2.6576269809660555,
     "sustainable": true,
-    "buffer_seconds": 60.754058678904,
+    "buffer_seconds": 60.020366739132946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357590.7798999976,
-    "upload_bytes_per_second": 4034032.713175629,
+    "generation_bytes_per_second": 316712.86506809364,
+    "upload_bytes_per_second": 4025277.3964436664,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.7306570792193,
+    "estimated_full_shards_per_hour": 67.9592081454478,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 1056341,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.5977993900014553,
+      "bytes": 5977865,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"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}
+{"step": 808200, "loss": 0.5658459067344666, "lr": 3e-05, "grad_norm": 0.3018091461766138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118296997231, "tokens_per_second": 150670.96458273006, "data_wait_s": 1.4690000170958228e-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 sha256:9d5df0f318e25a2bda0ba5c8bf4133cf8e4ccbd2cde951310d337db7c1218dc6
-size 172506
+oid sha256:095cfb7fa7bedded136e43a0d6f3d6d30876d69bb5bc41f0084f2629f82c7abb
+size 174624