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": 76978.20559941231,
-      "required_pages_per_second_upper_estimate": 0.005325530877412547
+      "required_token_positions_per_second": 76913.46739102445,
+      "required_pages_per_second_upper_estimate": 0.005321052137942936
     },
     {
       "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": 9622.275699926538,
-      "required_pages_per_second_upper_estimate": 0.004028645899593437
+      "required_token_positions_per_second": 9614.183423878056,
+      "required_pages_per_second_upper_estimate": 0.004025257832597877
     },
     {
       "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": 19244.551399853077,
-      "required_pages_per_second_upper_estimate": 0.004008132543640214
+      "required_token_positions_per_second": 19228.36684775611,
+      "required_pages_per_second_upper_estimate": 0.0040047617282537576
     },
     {
       "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": 28866.827099779613,
-      "required_pages_per_second_upper_estimate": 0.001730522207538953
+      "required_token_positions_per_second": 28842.55027163417,
+      "required_pages_per_second_upper_estimate": 0.0017290668487602039
     },
     {
       "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": 28866.827099779613,
-      "required_pages_per_second_upper_estimate": 0.0022442688352542414
+      "required_token_positions_per_second": 28842.55027163417,
+      "required_pages_per_second_upper_estimate": 0.0022423814186484153
     },
     {
       "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": 28866.827099779613,
-      "required_pages_per_second_upper_estimate": 0.3702726632518774
+      "required_token_positions_per_second": 28842.55027163417,
+      "required_pages_per_second_upper_estimate": 0.3699612661668548
     }
   ],
   "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": 0.07531253214050082,
-    "producer_p95_seconds": 0.20823064099931798,
-    "gpu_update_seconds": 0.9577775868623636,
-    "producer_batches_per_second": 13.278002632210397,
-    "gpu_batches_per_second": 1.0440837348010565,
-    "producer_headroom": 12.717373319430585,
+    "producer_batch_seconds": 0.055066561726576424,
+    "producer_p95_seconds": 0.18290303899993887,
+    "gpu_update_seconds": 0.9585837500365224,
+    "producer_batches_per_second": 18.159840902457805,
+    "gpu_batches_per_second": 1.0432056666534348,
+    "producer_headroom": 17.407728392344627,
     "sustainable": true,
-    "buffer_seconds": 60.339987972328906,
+    "buffer_seconds": 60.39077625230091,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376317.78805254283,
-    "upload_bytes_per_second": 5548421.434368912,
+    "generation_bytes_per_second": 376265.8282271124,
+    "upload_bytes_per_second": 5511079.103110042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.74903708631719,
+    "estimated_full_shards_per_hour": 80.73788771734266,
     "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": 1.4740382569998474,
-      "bytes": 11539917,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.530777683999986,
       "bytes": 10014604,
@@ -562,6 +556,12 @@
       "bytes": 1134238,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0769621990002634,
+      "bytes": 6468284,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 771800, "loss": 0.5233267545700073, "lr": 3e-05, "grad_norm": 0.2515248202104235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112969288643, "tokens_per_second": 151612.87926419108, "data_wait_s": 1.1820999134215526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 771900, "loss": 0.4768349826335907, "lr": 3e-05, "grad_norm": 0.23482950535474612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112983789567, "tokens_per_second": 151369.05276437572, "data_wait_s": 1.1676998838083819e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 772000, "loss": 0.5808287858963013, "lr": 3e-05, "grad_norm": 0.2571307187299208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112998450793, "tokens_per_second": 152753.22543473763, "data_wait_s": 9.157000022241846e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 772100, "loss": 0.6455023884773254, "lr": 3e-05, "grad_norm": 0.32084458493375906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113013083429, "tokens_per_second": 153113.03166792434, "data_wait_s": 1.0337000276194885e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f49d59d01e63e67257179b8235c83bf4ce642d941c4f99e12dfd3c7d036e79e1
-size 255891
+oid sha256:9f5d1daf768c42ad63eec2d3053305e01ed8a78165f3c6f2fb5bb4d8b52e97ef
+size 258009