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": 76009.86494752223,
-      "required_pages_per_second_upper_estimate": 0.006588342448409766
+      "required_token_positions_per_second": 75157.77262950977,
+      "required_pages_per_second_upper_estimate": 0.006514485245892682
     },
     {
       "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": 9501.23311844028,
-      "required_pages_per_second_upper_estimate": 0.0037901839470401624
+      "required_token_positions_per_second": 9394.721578688721,
+      "required_pages_per_second_upper_estimate": 0.003747694901343833
     },
     {
       "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": 19002.46623688056,
-      "required_pages_per_second_upper_estimate": 0.004395886131918819
+      "required_token_positions_per_second": 18789.443157377442,
+      "required_pages_per_second_upper_estimate": 0.004346606991553925
     },
     {
       "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": 28503.699355320838,
-      "required_pages_per_second_upper_estimate": 0.00173572574338615
+      "required_token_positions_per_second": 28184.164736066163,
+      "required_pages_per_second_upper_estimate": 0.0017162677615421233
     },
     {
       "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": 28503.699355320838,
-      "required_pages_per_second_upper_estimate": 0.0021942936796624487
+      "required_token_positions_per_second": 28184.164736066163,
+      "required_pages_per_second_upper_estimate": 0.0021696950201437856
     },
     {
       "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": 28503.699355320838,
-      "required_pages_per_second_upper_estimate": 0.353920550247971
+      "required_token_positions_per_second": 28184.164736066163,
+      "required_pages_per_second_upper_estimate": 0.34995299969040516
     }
   ],
   "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": 0.12869377285937844,
-    "producer_p95_seconds": 0.37103918499997235,
-    "gpu_update_seconds": 0.9699793579544228,
-    "producer_batches_per_second": 7.7703837394889606,
-    "gpu_batches_per_second": 1.030949774136315,
-    "producer_headroom": 7.537111830688989,
+    "producer_batch_seconds": 0.17772748179689302,
+    "producer_p95_seconds": 0.3545032580004772,
+    "gpu_update_seconds": 0.9809763836861182,
+    "producer_batches_per_second": 5.626591846627299,
+    "gpu_batches_per_second": 1.0193925324098003,
+    "producer_headroom": 5.519553722182246,
     "sustainable": true,
-    "buffer_seconds": 60.138720193174215,
+    "buffer_seconds": 60.820535788539324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329599.482107346,
-    "upload_bytes_per_second": 5353449.963755818,
+    "generation_bytes_per_second": 369529.9183429277,
+    "upload_bytes_per_second": 5353222.851677012,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.72437617697987,
+    "estimated_full_shards_per_hour": 79.29251826015351,
     "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": 1.0871770170015225,
-      "bytes": 6926231,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.2097226180012512,
       "bytes": 7368748,
@@ -562,6 +556,12 @@
       "bytes": 1164850,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.6704007100015588,
+      "bytes": 10030549,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -285,3 +285,4 @@
 {"step": 143500, "loss": 0.592973381280899, "lr": 0.0003, "grad_norm": 0.14191737250964032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21001089296, "tokens_per_second": 149698.6787272301, "data_wait_s": 1.6450001567136496e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143600, "loss": 0.5629896000027657, "lr": 0.0003, "grad_norm": 0.13351216367192972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21015732926, "tokens_per_second": 149379.60506217225, "data_wait_s": 1.227200118592009e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143700, "loss": 0.6280402727425098, "lr": 0.0003, "grad_norm": 0.15036053841949226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21030474349, "tokens_per_second": 149733.17884332212, "data_wait_s": 1.2964002962689847e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 143800, "loss": 0.5547901634126902, "lr": 0.0003, "grad_norm": 0.13474167818185723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21045143923, "tokens_per_second": 150466.8883624078, "data_wait_s": 1.4326000382425264e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42c7d55d673a8fe17e638dc05e4c34e7d22d308979d3c384cf2b1bafe8de4f64
-size 611946
+oid sha256:491c875a732ea9bfce1171fd13f0fc738b5fb98cc350eaae9c8e0cc1102afd9e
+size 614064