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": 77049.96630814987,
-      "required_pages_per_second_upper_estimate": 0.005330495449750797
+      "required_token_positions_per_second": 77448.80111295018,
+      "required_pages_per_second_upper_estimate": 0.005358087766971127
     },
     {
       "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": 9631.245788518734,
-      "required_pages_per_second_upper_estimate": 0.004032401488370239
+      "required_token_positions_per_second": 9681.100139118773,
+      "required_pages_per_second_upper_estimate": 0.00405327446388925
     },
     {
       "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": 19262.49157703747,
-      "required_pages_per_second_upper_estimate": 0.004011869009433435
+      "required_token_positions_per_second": 19362.200278237546,
+      "required_pages_per_second_upper_estimate": 0.004032635702398134
     },
     {
       "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": 28893.7373655562,
-      "required_pages_per_second_upper_estimate": 0.0017321354368826627
+      "required_token_positions_per_second": 29043.30041735632,
+      "required_pages_per_second_upper_estimate": 0.0017411015134685228
     },
     {
       "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": 28893.7373655562,
-      "required_pages_per_second_upper_estimate": 0.0022463609900526212
+      "required_token_positions_per_second": 29043.30041735632,
+      "required_pages_per_second_upper_estimate": 0.0022579888594717406
     },
     {
       "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": 28893.7373655562,
-      "required_pages_per_second_upper_estimate": 0.3706178392472672
+      "required_token_positions_per_second": 29043.30041735632,
+      "required_pages_per_second_upper_estimate": 0.37253627348746576
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06667956513285844,
-    "producer_p95_seconds": 0.20637315000021772,
-    "gpu_update_seconds": 0.9568855579395822,
-    "producer_batches_per_second": 14.997098406498436,
-    "gpu_batches_per_second": 1.0450570517055917,
-    "producer_headroom": 14.350506876177075,
+    "producer_batch_seconds": 0.1649795926250306,
+    "producer_p95_seconds": 0.20735765999961586,
+    "gpu_update_seconds": 0.951957924984225,
+    "producer_batches_per_second": 6.0613557355110155,
+    "gpu_batches_per_second": 1.0504665949564642,
+    "producer_headroom": 5.770155628568298,
     "sustainable": true,
-    "buffer_seconds": 60.28379015019368,
+    "buffer_seconds": 60.9253071989904,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393828.15701383096,
-    "upload_bytes_per_second": 3615412.57026016,
+    "generation_bytes_per_second": 400827.6627780786,
+    "upload_bytes_per_second": 3684962.1192163355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.50635464488217,
+    "estimated_full_shards_per_hour": 86.00828564173477,
     "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": 1.26191803599977,
-      "bytes": 2574745,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4942966790003993,
       "bytes": 10212337,
@@ -562,6 +556,12 @@
       "bytes": 7415076,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.515185850999842,
+      "bytes": 11700777,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -161,3 +161,4 @@
 {"step": 941100, "loss": 0.5392598509788513, "lr": 3e-05, "grad_norm": 0.36996917852931727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137750354229, "tokens_per_second": 153107.729645144, "data_wait_s": 1.302700002270285e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941200, "loss": 0.5398966670036316, "lr": 3e-05, "grad_norm": 0.3430767914103548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137765008400, "tokens_per_second": 153301.6113073163, "data_wait_s": 1.3201999536249787e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941300, "loss": 0.500248908996582, "lr": 3e-05, "grad_norm": 0.3266684065896893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137779724167, "tokens_per_second": 153474.19305677293, "data_wait_s": 1.4176999684423208e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 941400, "loss": 0.5718052387237549, "lr": 3e-05, "grad_norm": 0.3593922088571367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137794430820, "tokens_per_second": 147108.89537589918, "data_wait_s": 1.6684000001987442e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61d6cd5c2b6ec5f22c45e4238b66f7d83b54f040e9e0cc3654dc667c568ec598
-size 347748
+oid sha256:9bd37616af54f750dd273a68a24360472a0966d3a3c2cdc148f316be4f9ea737
+size 349866