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": 75233.1545221625,
-      "required_pages_per_second_upper_estimate": 0.00652101915729416
+      "required_token_positions_per_second": 75612.39009080868,
+      "required_pages_per_second_upper_estimate": 0.0065538903352206496
     },
     {
       "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": 9404.144315270312,
-      "required_pages_per_second_upper_estimate": 0.0037514537718486966
+      "required_token_positions_per_second": 9451.548761351085,
+      "required_pages_per_second_upper_estimate": 0.003770364114149946
     },
     {
       "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": 18808.288630540625,
-      "required_pages_per_second_upper_estimate": 0.004350966560101227
+      "required_token_positions_per_second": 18903.09752270217,
+      "required_pages_per_second_upper_estimate": 0.0043728989287232884
     },
     {
       "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": 28212.432945810935,
-      "required_pages_per_second_upper_estimate": 0.0017179891472037456
+      "required_token_positions_per_second": 28354.64628405326,
+      "required_pages_per_second_upper_estimate": 0.0017266491933669817
     },
     {
       "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": 28212.432945810935,
-      "required_pages_per_second_upper_estimate": 0.002171871185181351
+      "required_token_positions_per_second": 28354.64628405326,
+      "required_pages_per_second_upper_estimate": 0.0021828191616309616
     },
     {
       "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": 28212.432945810935,
-      "required_pages_per_second_upper_estimate": 0.35030399624782316
+      "required_token_positions_per_second": 28354.64628405326,
+      "required_pages_per_second_upper_estimate": 0.35206980995136716
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08389255442173749,
-    "producer_p95_seconds": 0.2780043279999518,
-    "gpu_update_seconds": 0.9799934678836429,
-    "producer_batches_per_second": 11.920008955417966,
-    "gpu_batches_per_second": 1.0204149647645737,
-    "producer_headroom": 11.681530913424133,
+    "producer_batch_seconds": 0.09801093524191629,
+    "producer_p95_seconds": 0.2728236779985309,
+    "gpu_update_seconds": 0.9750782895694002,
+    "producer_batches_per_second": 10.202943146412611,
+    "gpu_batches_per_second": 1.02555867636188,
+    "producer_headroom": 9.948668351777844,
     "sustainable": true,
-    "buffer_seconds": 60.759595008785865,
+    "buffer_seconds": 60.454853953302816,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326366.0086389438,
-    "upload_bytes_per_second": 4859561.811817777,
+    "generation_bytes_per_second": 353155.38748442975,
+    "upload_bytes_per_second": 4905403.596303916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.03054804207072,
+    "estimated_full_shards_per_hour": 75.77892511748952,
     "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": 0.8769548000000214,
-      "bytes": 1135557,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 3.838755348999257,
       "bytes": 11603941,
@@ -562,6 +556,12 @@
       "bytes": 13345421,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.0539056280031218,
+      "bytes": 6000362,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 254400, "loss": 0.5483874548226595, "lr": 0.0003, "grad_norm": 0.14924686155782332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37237388209, "tokens_per_second": 149804.04372506248, "data_wait_s": 7.629001629538834e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 254500, "loss": 0.553221432492137, "lr": 0.0003, "grad_norm": 0.14108443836552215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37251860553, "tokens_per_second": 147419.75755384032, "data_wait_s": 1.4804001693846658e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 254600, "loss": 0.5988007672131062, "lr": 0.0003, "grad_norm": 0.17240488792494962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37266504541, "tokens_per_second": 149974.5026438018, "data_wait_s": 1.3975000911159441e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 254700, "loss": 0.5812123045325279, "lr": 0.0003, "grad_norm": 0.13361758288913533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37281140164, "tokens_per_second": 149668.77361416118, "data_wait_s": 7.752001693006605e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc1a943ac934a1cb12d81dfada05664e7101a9c89bce32c89e51713d1ab66725
-size 373164
+oid sha256:00297cfe2085e301d0d37b900cd167bf84848a1016654768fcd1cffc504e21a2
+size 375282