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": 76562.81705670696,
-      "required_pages_per_second_upper_estimate": 0.005296793334193982
+      "required_token_positions_per_second": 76943.23976954863,
+      "required_pages_per_second_upper_estimate": 0.005323111860170674
     },
     {
       "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": 9570.35213208837,
-      "required_pages_per_second_upper_estimate": 0.004006906586026991
+      "required_token_positions_per_second": 9617.904971193579,
+      "required_pages_per_second_upper_estimate": 0.004026815966744141
     },
     {
       "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": 19140.70426417674,
-      "required_pages_per_second_upper_estimate": 0.003986503923911966
+      "required_token_positions_per_second": 19235.809942387157,
+      "required_pages_per_second_upper_estimate": 0.004006311928577799
     },
     {
       "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": 28711.05639626511,
-      "required_pages_per_second_upper_estimate": 0.0017211839917113509
+      "required_token_positions_per_second": 28853.714913580738,
+      "required_pages_per_second_upper_estimate": 0.0017297361519974794
     },
     {
       "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": 28711.05639626511,
-      "required_pages_per_second_upper_estimate": 0.0022321583482189035
+      "required_token_positions_per_second": 28853.714913580738,
+      "required_pages_per_second_upper_estimate": 0.002243249420451691
     },
     {
       "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": 28711.05639626511,
-      "required_pages_per_second_upper_estimate": 0.36827460392074385
+      "required_token_positions_per_second": 28853.714913580738,
+      "required_pages_per_second_upper_estimate": 0.3701044742060869
     }
   ],
   "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": 0.09275335117189343,
-    "producer_p95_seconds": 0.19160169399947335,
-    "gpu_update_seconds": 0.9629739713651428,
-    "producer_batches_per_second": 10.781281618027672,
-    "gpu_batches_per_second": 1.0384496671102832,
-    "producer_headroom": 10.38209357611812,
+    "producer_batch_seconds": 0.054609219625014305,
+    "producer_p95_seconds": 0.17371433400057867,
+    "gpu_update_seconds": 0.9582128361220746,
+    "producer_batches_per_second": 18.311926207089396,
+    "gpu_batches_per_second": 1.0436094803812477,
+    "producer_headroom": 17.546722745753275,
     "sustainable": true,
-    "buffer_seconds": 60.667360196004,
+    "buffer_seconds": 60.367408675690704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 275038.44143234764,
-    "upload_bytes_per_second": 5269746.280830422,
+    "generation_bytes_per_second": 274707.2431902458,
+    "upload_bytes_per_second": 5273935.104153618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.016846964147774,
+    "estimated_full_shards_per_hour": 58.94577953129321,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 8293432,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1932894399997167,
+      "bytes": 6531618,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 891700, "loss": 0.5582869052886963, "lr": 3e-05, "grad_norm": 0.33334995990102567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130519945362, "tokens_per_second": 146788.8785535297, "data_wait_s": 1.2253999557287898e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 891800, "loss": 0.5598562359809875, "lr": 3e-05, "grad_norm": 0.3153587736756321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130534588770, "tokens_per_second": 152931.06857877207, "data_wait_s": 9.547000445309095e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 891900, "loss": 0.5048050880432129, "lr": 3e-05, "grad_norm": 0.3236841189460622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130549141147, "tokens_per_second": 152273.73150246788, "data_wait_s": 9.749000128067564e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 892000, "loss": 0.5161802172660828, "lr": 3e-05, "grad_norm": 0.30836670864348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130563705809, "tokens_per_second": 152269.44595199273, "data_wait_s": 1.0167000255023595e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:daddd799d582c293a3bd4458b780a4a6a701fbd5ab05b081c6b65bb91e97f66d
-size 147090
+oid sha256:6d07a31084e650c6ca1fd9fed3a70edb2c1760cb709727b1e8de50fdda78a9f0
+size 149208