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": 75139.92167780774,
-      "required_pages_per_second_upper_estimate": 0.006512937970641973
+      "required_token_positions_per_second": 74931.9128957301,
+      "required_pages_per_second_upper_estimate": 0.006494908296604922
     },
     {
       "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": 9392.490209725967,
-      "required_pages_per_second_upper_estimate": 0.003746804774902652
+      "required_token_positions_per_second": 9366.489111966262,
+      "required_pages_per_second_upper_estimate": 0.0037364325482552507
     },
     {
       "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": 18784.980419451935,
-      "required_pages_per_second_upper_estimate": 0.004345574615676364
+      "required_token_positions_per_second": 18732.978223932525,
+      "required_pages_per_second_upper_estimate": 0.004333544822950331
     },
     {
       "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": 28177.470629177904,
-      "required_pages_per_second_upper_estimate": 0.0017158601255538928
+      "required_token_positions_per_second": 28099.46733589879,
+      "required_pages_per_second_upper_estimate": 0.001711110134244846
     },
     {
       "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": 28177.470629177904,
-      "required_pages_per_second_upper_estimate": 0.0021691796892650537
+      "required_token_positions_per_second": 28099.46733589879,
+      "required_pages_per_second_upper_estimate": 0.0021631747798214916
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28177.470629177904,
-      "required_pages_per_second_upper_estimate": 0.3498698812865876
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28099.46733589879,
+      "required_pages_per_second_upper_estimate": 0.3489013414442901
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2983639846171826,
-    "producer_p95_seconds": 0.2563801289998082,
-    "gpu_update_seconds": 0.9812094337300228,
-    "producer_batches_per_second": 3.3516109569425914,
-    "gpu_batches_per_second": 1.019150413381724,
-    "producer_headroom": 3.2886322891449797,
+    "producer_batch_seconds": 0.06344004382042101,
+    "producer_p95_seconds": 0.21984489300120913,
+    "gpu_update_seconds": 0.9839332422034203,
+    "producer_batches_per_second": 15.762914710946422,
+    "gpu_batches_per_second": 1.016329113711617,
+    "producer_headroom": 15.509655778117502,
     "sustainable": true,
-    "buffer_seconds": 60.83498489126141,
+    "buffer_seconds": 60.019927774408636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450951.7568081669,
-    "upload_bytes_per_second": 5229687.106816401,
+    "generation_bytes_per_second": 450979.9068367697,
+    "upload_bytes_per_second": 5288226.860244146,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.76374939140086,
+    "estimated_full_shards_per_hour": 96.7697897322399,
     "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": 1.0863462409997737,
-      "bytes": 6279325,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9579987879997134,
       "bytes": 7302822,
@@ -562,6 +556,12 @@
       "bytes": 2518854,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.2675222210000356,
+      "bytes": 11869039,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 206300, "loss": 0.5879179909825325, "lr": 0.0003, "grad_norm": 0.13954754362615665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30194524871, "tokens_per_second": 149234.82555407833, "data_wait_s": 1.6312000298057683e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206400, "loss": 0.6384740583598614, "lr": 0.0003, "grad_norm": 0.17451350463504176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30209255939, "tokens_per_second": 148678.38793833062, "data_wait_s": 2.5233000997104682e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206500, "loss": 0.5908651724457741, "lr": 0.0003, "grad_norm": 0.1655434847203432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30223869542, "tokens_per_second": 147681.66956276508, "data_wait_s": 1.7620999642531388e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 206600, "loss": 0.5545372609049082, "lr": 0.0003, "grad_norm": 0.1425858499947372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30238319105, "tokens_per_second": 145724.7558609718, "data_wait_s": 1.7537000530865043e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:34fac9a54bc856d0be8aee566f96f5a7702fbd424ddbbc985599b8ab10530368
-size 245301
+oid sha256:eae82839c5a34e78092f57ecf7a454c42fad876af6748538ebd02c354372652a
+size 247419