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": 77034.66708698198,
-      "required_pages_per_second_upper_estimate": 0.005329437014131318
+      "required_token_positions_per_second": 77245.81026749732,
+      "required_pages_per_second_upper_estimate": 0.005344044389279046
     },
     {
       "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": 9629.333385872747,
-      "required_pages_per_second_upper_estimate": 0.004031600805317873
+      "required_token_positions_per_second": 9655.726283437165,
+      "required_pages_per_second_upper_estimate": 0.00404265096038172
     },
     {
       "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": 19258.666771745495,
-      "required_pages_per_second_upper_estimate": 0.0040110724033580155
+      "required_token_positions_per_second": 19311.45256687433,
+      "required_pages_per_second_upper_estimate": 0.004022066292428322
     },
     {
       "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": 28888.000157618244,
-      "required_pages_per_second_upper_estimate": 0.0017317915000269915
+      "required_token_positions_per_second": 28967.178850311495,
+      "required_pages_per_second_upper_estimate": 0.0017365381417550883
     },
     {
       "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": 28888.000157618244,
-      "required_pages_per_second_upper_estimate": 0.002245914947370756
+      "required_token_positions_per_second": 28967.178850311495,
+      "required_pages_per_second_upper_estimate": 0.0022520707424573934
     },
     {
       "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": 28888.000157618244,
-      "required_pages_per_second_upper_estimate": 0.37054424850397305
+      "required_token_positions_per_second": 28967.178850311495,
+      "required_pages_per_second_upper_estimate": 0.37155986775838556
     }
   ],
   "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": 0.2571736468203234,
-    "producer_p95_seconds": 0.15720274600016637,
-    "gpu_update_seconds": 0.9570755971042448,
-    "producer_batches_per_second": 3.8884232982808644,
-    "gpu_batches_per_second": 1.0448495427379283,
-    "producer_headroom": 3.7215150499962153,
+    "producer_batch_seconds": 0.04664854805465879,
+    "producer_p95_seconds": 0.1546057049999945,
+    "gpu_update_seconds": 0.9544595330760935,
+    "producer_batches_per_second": 21.436894430846706,
+    "gpu_batches_per_second": 1.0477133554076785,
+    "producer_headroom": 20.460648249067457,
     "sustainable": true,
-    "buffer_seconds": 60.295762617567426,
+    "buffer_seconds": 60.13095058379389,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400449.63492629747,
-    "upload_bytes_per_second": 24070.443584197794,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 398665.78927728994,
+    "upload_bytes_per_second": 4796886.720763712,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.92716966478055,
+    "estimated_full_shards_per_hour": 85.54439791430495,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 23380,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.497052147999966,
+      "bytes": 11817101,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -36,3 +36,4 @@
 {"step": 763600, "loss": 0.5196139216423035, "lr": 3e-05, "grad_norm": 0.25202850827053114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111768630639, "tokens_per_second": 153365.36281048146, "data_wait_s": 1.4491999991150806e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 763700, "loss": 0.5006608963012695, "lr": 3e-05, "grad_norm": 0.24746026926897136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111783264294, "tokens_per_second": 152010.29752294873, "data_wait_s": 1.3126999874657486e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 763800, "loss": 0.5564277172088623, "lr": 3e-05, "grad_norm": 0.24487768077464622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111797752354, "tokens_per_second": 150539.70673072373, "data_wait_s": 1.1176999578310642e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 763900, "loss": 0.5075767040252686, "lr": 3e-05, "grad_norm": 0.23713249721745336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111812350458, "tokens_per_second": 151700.1328655073, "data_wait_s": 1.1004000043612905e-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:a90eb63cbeae2ed7ee3c2cb3c6254c2a41e32499a08950b18a1cc572b94cf3d1
-size 80649
+oid sha256:47ab9d8c0b69668a85c9fa830ae0ab8b97ff94f092496ec294023644f75c4bbc
+size 82767