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": 74826.2429459554,
-      "required_pages_per_second_upper_estimate": 0.00648574909317649
+      "required_token_positions_per_second": 74770.47765798795,
+      "required_pages_per_second_upper_estimate": 0.006480915499351302
     },
     {
       "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": 9353.280368244425,
-      "required_pages_per_second_upper_estimate": 0.003731163382896292
+      "required_token_positions_per_second": 9346.309707248494,
+      "required_pages_per_second_upper_estimate": 0.003728382682004346
     },
     {
       "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": 18706.56073648885,
-      "required_pages_per_second_upper_estimate": 0.0043274336021621364
+      "required_token_positions_per_second": 18692.61941449699,
+      "required_pages_per_second_upper_estimate": 0.004324208522678207
     },
     {
       "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": 28059.841104733274,
-      "required_pages_per_second_upper_estimate": 0.0017086971046696333
+      "required_token_positions_per_second": 28038.929121745485,
+      "required_pages_per_second_upper_estimate": 0.0017074236746223725
     },
     {
       "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": 28059.841104733274,
-      "required_pages_per_second_upper_estimate": 0.0021601242428539416
+      "required_token_positions_per_second": 28038.929121745485,
+      "required_pages_per_second_upper_estimate": 0.002158514380515468
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28059.841104733274,
-      "required_pages_per_second_upper_estimate": 0.3484093162736788
+      "required_token_positions_per_second": 28038.929121745485,
+      "required_pages_per_second_upper_estimate": 0.3481496594328754
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.097586275515539,
-    "producer_p95_seconds": 0.20200520000071265,
-    "gpu_update_seconds": 0.9853227570606662,
-    "producer_batches_per_second": 10.247342617771764,
-    "gpu_batches_per_second": 1.0148958732904108,
-    "producer_headroom": 10.096939880688138,
+    "producer_batch_seconds": 0.05695632236715653,
+    "producer_p95_seconds": 0.1906214450009429,
+    "gpu_update_seconds": 0.9860576300882226,
+    "producer_batches_per_second": 17.55731336643749,
+    "gpu_batches_per_second": 1.0141395081649842,
+    "producer_headroom": 17.312522808825626,
     "sustainable": true,
-    "buffer_seconds": 60.104688180700634,
+    "buffer_seconds": 60.14951543538157,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301655.57109729195,
-    "upload_bytes_per_second": 5783006.343312844,
+    "generation_bytes_per_second": 342675.44495674816,
+    "upload_bytes_per_second": 5844408.80102132,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.72826337517805,
+    "estimated_full_shards_per_hour": 73.53017341162523,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 1195413,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1174536390008143,
+      "bytes": 7647354,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 91700, "loss": 0.6243661977350712, "lr": 0.0003, "grad_norm": 0.1397994986921189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13419543507, "tokens_per_second": 148050.62269793818, "data_wait_s": 1.1871999959112145e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91800, "loss": 0.5888327732682228, "lr": 0.0003, "grad_norm": 0.12957206124342208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13434264611, "tokens_per_second": 150189.06462793087, "data_wait_s": 1.0155999916605651e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91900, "loss": 0.5598646011203527, "lr": 0.0003, "grad_norm": 0.1423399720042594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13449024303, "tokens_per_second": 150284.03344551637, "data_wait_s": 1.3726999895879999e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 92000, "loss": 0.6390262767672539, "lr": 0.0003, "grad_norm": 0.1441192939934209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13463728262, "tokens_per_second": 149921.50896801552, "data_wait_s": 9.022000085678883e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e728614224c9cc26f4ef0b422458c75b8d991cf2ad0636b038f1a681e608477b
-size 147090
+oid sha256:b302d39b6f302b7d81b1e00f9d0069889461bcd015ae8830bc027787ba1022ec
+size 149208