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": 75375.59441535952,
-      "required_pages_per_second_upper_estimate": 0.006533365486225869
+      "required_token_positions_per_second": 75109.98073137106,
+      "required_pages_per_second_upper_estimate": 0.006510342765289477
     },
     {
       "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": 9421.94930191994,
-      "required_pages_per_second_upper_estimate": 0.0037585564472315063
+      "required_token_positions_per_second": 9388.747591421383,
+      "required_pages_per_second_upper_estimate": 0.003745311788503823
     },
     {
       "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": 18843.89860383988,
-      "required_pages_per_second_upper_estimate": 0.004359204300709885
+      "required_token_positions_per_second": 18777.495182842766,
+      "required_pages_per_second_upper_estimate": 0.004343843037922498
     },
     {
       "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": 28265.84790575982,
-      "required_pages_per_second_upper_estimate": 0.0017212418380179962
+      "required_token_positions_per_second": 28166.24277426415,
+      "required_pages_per_second_upper_estimate": 0.0017151764080976472
     },
     {
       "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": 28265.84790575982,
-      "required_pages_per_second_upper_estimate": 0.002175983216660289
+      "required_token_positions_per_second": 28166.24277426415,
+      "required_pages_per_second_upper_estimate": 0.002168315337913101
     },
     {
       "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": 28265.84790575982,
-      "required_pages_per_second_upper_estimate": 0.3509672312820172
+      "required_token_positions_per_second": 28166.24277426415,
+      "required_pages_per_second_upper_estimate": 0.34973046890577186
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.3000724374843742,
-    "producer_p95_seconds": 0.29751322800075286,
-    "gpu_update_seconds": 0.9781415400019216,
-    "producer_batches_per_second": 3.3325286666892673,
-    "gpu_batches_per_second": 1.022346929461799,
-    "producer_headroom": 3.25968472213599,
+    "producer_batch_seconds": 0.4886274724141657,
+    "producer_p95_seconds": 0.30333658300151,
+    "gpu_update_seconds": 0.9816005713499824,
+    "producer_batches_per_second": 2.046548866888904,
+    "gpu_batches_per_second": 1.0187443133052716,
+    "producer_headroom": 2.0088935370338072,
     "sustainable": true,
-    "buffer_seconds": 60.64477548011914,
+    "buffer_seconds": 60.859235423698905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 230295.36451009245,
-    "upload_bytes_per_second": 6274388.174995325,
+    "generation_bytes_per_second": 211697.35794035607,
+    "upload_bytes_per_second": 6273618.576206602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.41602422215538,
+    "estimated_full_shards_per_hour": 45.425324951725116,
     "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": 1.0259071060008864,
-      "bytes": 6033717,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.3882136779993743,
       "bytes": 13040894,
@@ -562,6 +556,12 @@
       "bytes": 6189988,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1359487030003947,
+      "bytes": 6670820,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 15000, "loss": 0.6725511886179447, "lr": 0.0003, "grad_norm": 0.14628862923077823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2194710862, "tokens_per_second": 127660.50463823703, "data_wait_s": 1.1482999980216846e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 15100, "loss": 0.6722135487943888, "lr": 0.0003, "grad_norm": 0.15125409810689924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2209309355, "tokens_per_second": 116584.98687357442, "data_wait_s": 7.790999006829225e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 15200, "loss": 0.6872422629967332, "lr": 0.0003, "grad_norm": 0.142948339273216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2223893021, "tokens_per_second": 148380.3749724633, "data_wait_s": 1.2586000593728386e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 15300, "loss": 0.6712484434247017, "lr": 0.0003, "grad_norm": 0.14542909980272553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2238395506, "tokens_per_second": 130379.01008881455, "data_wait_s": 1.3466999007505365e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:609973f428cf3ce634d81f52acd5a9b17b70d09d8a1768d5d29ae8e02b4ea7ac
-size 319061
+oid sha256:ab6b07f33929e0f0ef182da1ae5936fc8feaf6753195f49b16e5acd5cb336e4a
+size 321144