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": 75267.63065437644,
-      "required_pages_per_second_upper_estimate": 0.0065240074610555055
+      "required_token_positions_per_second": 75321.02865676544,
+      "required_pages_per_second_upper_estimate": 0.006528635864566585
     },
     {
       "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": 9408.453831797055,
-      "required_pages_per_second_upper_estimate": 0.0037531729024242277
+      "required_token_positions_per_second": 9415.12858209568,
+      "required_pages_per_second_upper_estimate": 0.00375583556011476
     },
     {
       "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": 18816.90766359411,
-      "required_pages_per_second_upper_estimate": 0.00435296042170835
+      "required_token_positions_per_second": 18830.25716419136,
+      "required_pages_per_second_upper_estimate": 0.004356048593728338
     },
     {
       "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": 28225.361495391164,
-      "required_pages_per_second_upper_estimate": 0.001718776428042324
+      "required_token_positions_per_second": 28245.38574628704,
+      "required_pages_per_second_upper_estimate": 0.0017199957998627578
     },
     {
       "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": 28225.361495391164,
-      "required_pages_per_second_upper_estimate": 0.002172866460716553
+      "required_token_positions_per_second": 28245.38574628704,
+      "required_pages_per_second_upper_estimate": 0.0021744079829811914
     },
     {
       "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": 28225.361495391164,
-      "required_pages_per_second_upper_estimate": 0.3504645255645376
+      "required_token_positions_per_second": 28245.38574628704,
+      "required_pages_per_second_upper_estimate": 0.3507131597438077
     }
   ],
   "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": 0.10579500924222884,
-    "producer_p95_seconds": 0.37422385800164193,
-    "gpu_update_seconds": 0.9795445845579183,
-    "producer_batches_per_second": 9.452241718797854,
-    "gpu_batches_per_second": 1.0208825772349235,
-    "producer_headroom": 9.258892187580868,
+    "producer_batch_seconds": 0.0905434923356836,
+    "producer_p95_seconds": 0.34721226500187186,
+    "gpu_update_seconds": 0.9788501473602438,
+    "producer_batches_per_second": 11.044416050273064,
+    "gpu_batches_per_second": 1.021606833994757,
+    "producer_headroom": 10.81082827831763,
     "sustainable": true,
-    "buffer_seconds": 60.731764242590934,
+    "buffer_seconds": 60.68870913633511,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304156.12709276844,
-    "upload_bytes_per_second": 4572950.767968387,
+    "generation_bytes_per_second": 306754.4923462239,
+    "upload_bytes_per_second": 4545911.852563372,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.2648244818429,
+    "estimated_full_shards_per_hour": 65.82237317838705,
     "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": 1.3431768080008624,
-      "bytes": 5992564,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2225116730005539,
       "bytes": 6709538,
@@ -562,6 +556,12 @@
       "bytes": 5881567,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8555704430000333,
+      "bytes": 1226801,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 184000, "loss": 0.6542017664760351, "lr": 0.0003, "grad_norm": 0.1794766250336593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26930024755, "tokens_per_second": 148902.9537748219, "data_wait_s": 1.353900006506592e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 184100, "loss": 0.5877811014652252, "lr": 0.0003, "grad_norm": 0.15011607646486622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26944726361, "tokens_per_second": 150477.16169769233, "data_wait_s": 1.326200072071515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 184200, "loss": 0.5851279124617577, "lr": 0.0003, "grad_norm": 0.1448492210517399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26959426724, "tokens_per_second": 150033.83465250413, "data_wait_s": 1.2533997505670413e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 184300, "loss": 0.6379616372287273, "lr": 0.0003, "grad_norm": 0.14567102136135143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26974147930, "tokens_per_second": 150217.05002835635, "data_wait_s": 1.0604002454783767e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9942cd3f2284573f189e0930ad7083a19667bb586a8a4de0b600c19bf16d3ae4
-size 622536
+oid sha256:1d1a6bc64b440a418b4ba846a8d6422ab9014b31e1ed55a8f237aefc0f64d971
+size 624654