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": 76262.06109115075,
-      "required_pages_per_second_upper_estimate": 0.006610202170954204
+      "required_token_positions_per_second": 76069.06281119368,
+      "required_pages_per_second_upper_estimate": 0.006593473569197194
     },
     {
       "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": 9532.757636393844,
-      "required_pages_per_second_upper_estimate": 0.0038027595485853856
+      "required_token_positions_per_second": 9508.63285139921,
+      "required_pages_per_second_upper_estimate": 0.003793135811153347
     },
     {
       "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": 19065.51527278769,
-      "required_pages_per_second_upper_estimate": 0.0044104714167508
+      "required_token_positions_per_second": 19017.26570279842,
+      "required_pages_per_second_upper_estimate": 0.004399309727897212
     },
     {
       "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": 28598.27290918153,
-      "required_pages_per_second_upper_estimate": 0.0017414847766271766
+      "required_token_positions_per_second": 28525.898554197633,
+      "required_pages_per_second_upper_estimate": 0.0017370775581275516
     },
     {
       "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": 28598.27290918153,
-      "required_pages_per_second_upper_estimate": 0.0022015742136349977
+      "required_token_positions_per_second": 28525.898554197633,
+      "required_pages_per_second_upper_estimate": 0.0021960026354433003
     },
     {
       "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": 28598.27290918153,
-      "required_pages_per_second_upper_estimate": 0.35509483726959695
+      "required_token_positions_per_second": 28525.898554197633,
+      "required_pages_per_second_upper_estimate": 0.3541961900020814
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09255114756234661,
-    "producer_p95_seconds": 0.3597117309982423,
-    "gpu_update_seconds": 0.966771667918574,
-    "producer_batches_per_second": 10.80483631309223,
-    "gpu_batches_per_second": 1.0343704032545404,
-    "producer_headroom": 10.44580962399535,
+    "producer_batch_seconds": 0.10617879225785032,
+    "producer_p95_seconds": 0.36039348499980406,
+    "gpu_update_seconds": 0.9692245082997237,
+    "producer_batches_per_second": 9.418076611490795,
+    "gpu_batches_per_second": 1.0317526965493935,
+    "producer_headroom": 9.128230672901294,
     "sustainable": true,
-    "buffer_seconds": 60.90661507887017,
+    "buffer_seconds": 60.09191951458287,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 532962.9461194361,
-    "upload_bytes_per_second": 5168622.69622346,
+    "generation_bytes_per_second": 546625.9737170503,
+    "upload_bytes_per_second": 5222114.950742523,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.36144149482071,
+    "estimated_full_shards_per_hour": 117.29320915826446,
     "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": 0.9678468759993848,
-      "bytes": 1151038,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.3402203920013562,
       "bytes": 12178587,
@@ -562,6 +556,12 @@
       "bytes": 932725,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1173031580001407,
+      "bytes": 6247577,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -317,3 +317,4 @@
 {"step": 146700, "loss": 0.6494546383619308, "lr": 0.0003, "grad_norm": 0.15985365485145217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21469065047, "tokens_per_second": 150763.3991742238, "data_wait_s": 1.048400008585304e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146800, "loss": 0.5871815280988812, "lr": 0.0003, "grad_norm": 0.1341965422469446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21483789927, "tokens_per_second": 151155.09506963321, "data_wait_s": 9.743998816702515e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 146900, "loss": 0.6355623006820679, "lr": 0.0003, "grad_norm": 0.1646902456067799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21498590125, "tokens_per_second": 150710.4243415569, "data_wait_s": 1.0867999662877992e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 147000, "loss": 0.5945109017193317, "lr": 0.0003, "grad_norm": 0.13763741233161803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21513190824, "tokens_per_second": 149468.14629271862, "data_wait_s": 9.603001672076061e-06, "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:d8bc4b1681900dca2c3f0ad6666b9964cf601ad96463de1822f0fe1ea3ef5394
-size 680505
+oid sha256:106f9b5e489f47045b5b13c65df6c1827c9bdacb06a1c8d91d7fe7ea7676f6e8
+size 682623