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": 74657.70714530139,
-      "required_pages_per_second_upper_estimate": 0.006471140837125882
+      "required_token_positions_per_second": 74598.66104826413,
+      "required_pages_per_second_upper_estimate": 0.006466022871085657
     },
     {
       "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": 9332.213393162674,
-      "required_pages_per_second_upper_estimate": 0.003722759451556835
+      "required_token_positions_per_second": 9324.832631033016,
+      "required_pages_per_second_upper_estimate": 0.0037198151551910866
     },
     {
       "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": 18664.426786325348,
-      "required_pages_per_second_upper_estimate": 0.004317686654324007
+      "required_token_positions_per_second": 18649.66526206603,
+      "required_pages_per_second_upper_estimate": 0.004314271835480568
     },
     {
       "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": 27996.640179488022,
-      "required_pages_per_second_upper_estimate": 0.0017048484999118258
+      "required_token_positions_per_second": 27974.497893099044,
+      "required_pages_per_second_upper_estimate": 0.0017035001508423355
     },
     {
       "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": 27996.640179488022,
-      "required_pages_per_second_upper_estimate": 0.0021552588606772037
+      "required_token_positions_per_second": 27974.497893099044,
+      "required_pages_per_second_upper_estimate": 0.0021535542861772085
     },
     {
       "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": 27996.640179488022,
-      "required_pages_per_second_upper_estimate": 0.3476245723020229
+      "required_token_positions_per_second": 27974.497893099044,
+      "required_pages_per_second_upper_estimate": 0.3473496392105373
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08412486164053234,
-    "producer_p95_seconds": 0.3134467819982092,
-    "gpu_update_seconds": 0.9875470707466013,
-    "producer_batches_per_second": 11.887092358891778,
-    "gpu_batches_per_second": 1.0126099601956027,
-    "producer_headroom": 11.73906323871788,
+    "producer_batch_seconds": 0.10667241947652428,
+    "producer_p95_seconds": 0.33940244700352196,
+    "gpu_update_seconds": 0.9883287308910167,
+    "producer_batches_per_second": 9.37449440921393,
+    "gpu_batches_per_second": 1.011809096249242,
+    "producer_headroom": 9.265082162203335,
     "sustainable": true,
-    "buffer_seconds": 60.24037131554268,
+    "buffer_seconds": 60.288052584352016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375129.03364450956,
-    "upload_bytes_per_second": 4151389.33518897,
+    "generation_bytes_per_second": 375113.9794113273,
+    "upload_bytes_per_second": 4074724.36375993,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.49395806313959,
+    "estimated_full_shards_per_hour": 80.49072777514328,
     "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": 1.1363001400022767,
-      "bytes": 9937212,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.7940713860007236,
       "bytes": 1067006,
@@ -562,6 +556,12 @@
       "bytes": 11720658,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.8457645430025877,
+      "bytes": 857794,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -250,3 +250,4 @@
 {"step": 110000, "loss": 0.5629305355250835, "lr": 0.0003, "grad_norm": 0.13700414221724075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16096459430, "tokens_per_second": 147489.31812914196, "data_wait_s": 1.3880002370569855e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110100, "loss": 0.5995736382901669, "lr": 0.0003, "grad_norm": 0.14184116113695153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16111112567, "tokens_per_second": 130778.05111079352, "data_wait_s": 2.0277999283280224e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110200, "loss": 0.6425034925341606, "lr": 0.0003, "grad_norm": 0.15780812135677313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16125762229, "tokens_per_second": 148640.36650387038, "data_wait_s": 1.4149998605716974e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 110300, "loss": 0.6462549883872271, "lr": 0.0003, "grad_norm": 0.14029656109056018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16140427131, "tokens_per_second": 148427.77492959303, "data_wait_s": 1.867600076366216e-05, "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:c7a8212ca9f4d7c9a169f3bd5e1e9cec28adba1b3ec8d5bfc4fd0cd7df2476b4
-size 537816
+oid sha256:339f78aad5e32ea36fdb1525a54bb868e5d546ac3c8ff6c9dd34c71c5503b598
+size 539934