Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75337.24122364733,
-      "required_pages_per_second_upper_estimate": 0.006530041128773532
+      "required_token_positions_per_second": 75089.9673637601,
+      "required_pages_per_second_upper_estimate": 0.006508608057308375
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9417.155152955916,
-      "required_pages_per_second_upper_estimate": 0.0037566439895308427
+      "required_token_positions_per_second": 9386.245920470012,
+      "required_pages_per_second_upper_estimate": 0.003744313834558007
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18834.310305911833,
-      "required_pages_per_second_upper_estimate": 0.0043569862167325115
+      "required_token_positions_per_second": 18772.491840940023,
+      "required_pages_per_second_upper_estimate": 0.0043426856028291085
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28251.465458867748,
-      "required_pages_per_second_upper_estimate": 0.0017203660224611361
+      "required_token_positions_per_second": 28158.737761410033,
+      "required_pages_per_second_upper_estimate": 0.001714719392190591
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28251.465458867748,
-      "required_pages_per_second_upper_estimate": 0.002174876016085376
+      "required_token_positions_per_second": 28158.737761410033,
+      "required_pages_per_second_upper_estimate": 0.002167737581248386
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28251.465458867748,
-      "required_pages_per_second_upper_estimate": 0.350788649426571
+      "required_token_positions_per_second": 28158.737761410033,
+      "required_pages_per_second_upper_estimate": 0.349637281763786
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12174314754673787,
-    "producer_p95_seconds": 0.31061179300013464,
-    "gpu_update_seconds": 0.978639498905062,
-    "producer_batches_per_second": 8.214014670649897,
-    "gpu_batches_per_second": 1.0218267310065012,
-    "producer_headroom": 8.038559201283643,
+    "producer_batch_seconds": 0.36885495375001653,
+    "producer_p95_seconds": 0.21972717700009525,
+    "gpu_update_seconds": 0.9818621926260498,
+    "producer_batches_per_second": 2.7110927746350084,
+    "gpu_batches_per_second": 1.0184728646343328,
+    "producer_headroom": 2.6619194961157704,
     "sustainable": true,
-    "buffer_seconds": 60.675648932113845,
+    "buffer_seconds": 60.87545594281509,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305806.19864395034,
-    "upload_bytes_per_second": 4032206.44387455,
+    "generation_bytes_per_second": 317340.0407211248,
+    "upload_bytes_per_second": 4045554.419808952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.61889142502673,
+    "estimated_full_shards_per_hour": 68.0937854406863,
     "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": 1.3994414539993159,
-      "bytes": 7586010,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0743332829997598,
       "bytes": 4951926,
@@ -562,6 +556,12 @@
       "bytes": 1077101,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.117781933000515,
+      "bytes": 7831651,
+      "pages": 22,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -136,3 +136,4 @@
 {"step": 328600, "loss": 0.5836680959910154, "lr": 0.0003, "grad_norm": 0.17039313687931704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48098041078, "tokens_per_second": 148264.05142376272, "data_wait_s": 1.5903999155852944e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 328700, "loss": 0.5688195750117302, "lr": 0.0003, "grad_norm": 0.17441126115072939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48112757984, "tokens_per_second": 149293.19647971992, "data_wait_s": 1.5879000784480013e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 328800, "loss": 0.660380631685257, "lr": 0.0003, "grad_norm": 0.20252718627519192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48127535874, "tokens_per_second": 148482.7347440647, "data_wait_s": 1.6338999557774514e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 328900, "loss": 0.6235296614468098, "lr": 0.0003, "grad_norm": 0.14611748385605103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48142190297, "tokens_per_second": 149227.0087026593, "data_wait_s": 1.6834999769343995e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e24b2eee991e6dc49ecc68ac5725e41c4fcd223a1405500b9cc65acec92a44e
-size 294015
+oid sha256:37fd62dced9fdc1ac1953f944a87f3aea85b4989abecd77fdfb4bff9f4e3b65d
+size 296133