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": 75764.72099614113,
-      "required_pages_per_second_upper_estimate": 0.0065670939920157165
+      "required_token_positions_per_second": 75093.12662598676,
+      "required_pages_per_second_upper_estimate": 0.006508881894151105
     },
     {
       "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": 9470.590124517641,
-      "required_pages_per_second_upper_estimate": 0.003777959998610835
+      "required_token_positions_per_second": 9386.640828248344,
+      "required_pages_per_second_upper_estimate": 0.0037444713691751814
     },
     {
       "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": 18941.180249035282,
-      "required_pages_per_second_upper_estimate": 0.004381708697227362
+      "required_token_positions_per_second": 18773.28165649669,
+      "required_pages_per_second_upper_estimate": 0.004342868312757864
     },
     {
       "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": 28411.77037355292,
-      "required_pages_per_second_upper_estimate": 0.001730127750710576
+      "required_token_positions_per_second": 28159.922484745035,
+      "required_pages_per_second_upper_estimate": 0.0017147915356259295
     },
     {
       "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": 28411.77037355292,
-      "required_pages_per_second_upper_estimate": 0.002187216732170246
+      "required_token_positions_per_second": 28159.922484745035,
+      "required_pages_per_second_upper_estimate": 0.0021678287845302386
     },
     {
       "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": 28411.77037355292,
-      "required_pages_per_second_upper_estimate": 0.3527790999609238
+      "required_token_positions_per_second": 28159.922484745035,
+      "required_pages_per_second_upper_estimate": 0.3496519920625928
     }
   ],
   "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": 0.07158776965621882,
-    "producer_p95_seconds": 0.2579312210000353,
-    "gpu_update_seconds": 0.9731178183016755,
-    "producer_batches_per_second": 13.96886653687122,
-    "gpu_batches_per_second": 1.02762479649714,
-    "producer_headroom": 13.593352928507402,
+    "producer_batch_seconds": 0.5556710595079437,
+    "producer_p95_seconds": 0.31581125000229804,
+    "gpu_update_seconds": 0.9818208844494385,
+    "producer_batches_per_second": 1.799625844983752,
+    "gpu_batches_per_second": 1.0185157148706971,
+    "producer_headroom": 1.7669102388000155,
     "sustainable": true,
-    "buffer_seconds": 60.33330473470389,
+    "buffer_seconds": 60.87289483586519,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 220005.6377702264,
-    "upload_bytes_per_second": 2221505.909325709,
+    "generation_bytes_per_second": 204903.86335895178,
+    "upload_bytes_per_second": 2241332.692021112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.208088396359386,
+    "estimated_full_shards_per_hour": 43.9675991590158,
     "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": 2.8297132219995547,
-      "bytes": 11437747,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1368421070001204,
       "bytes": 1161800,
@@ -562,6 +556,12 @@
       "bytes": 6033435,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.161550510001689,
+      "bytes": 11437647,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -171,3 +171,4 @@
 {"step": 297300, "loss": 0.6055130567401648, "lr": 0.0003, "grad_norm": 0.16041599723372443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43517132742, "tokens_per_second": 150014.6329969246, "data_wait_s": 1.1656000424409285e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 297400, "loss": 0.6196131110191345, "lr": 0.0003, "grad_norm": 0.1556306077614989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43531929409, "tokens_per_second": 151526.35165024063, "data_wait_s": 9.460996807320043e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 297500, "loss": 0.6662709899246693, "lr": 0.0003, "grad_norm": 0.1811783066483314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43546561769, "tokens_per_second": 132601.77143004382, "data_wait_s": 8.721999620320275e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 297600, "loss": 0.5978391785174608, "lr": 0.0003, "grad_norm": 0.15276766862988436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43560986804, "tokens_per_second": 147247.8215553319, "data_wait_s": 1.0168001608690247e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:117700b78528b144435c6781a4a698620a2d7149d057eb1688ab5e43d9f74e56
-size 368928
+oid sha256:ace7463fb8358573d6c4b6af00e9a9efdf2eae179c2f4704828fe66125639ec7
+size 371046