Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77507.31276979626,
-      "required_pages_per_second_upper_estimate": 0.005362135739157488
+      "required_token_positions_per_second": 77180.58715638207,
+      "required_pages_per_second_upper_estimate": 0.0053395321031136245
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9688.414096224533,
-      "required_pages_per_second_upper_estimate": 0.004056336664996639
+      "required_token_positions_per_second": 9647.573394547759,
+      "required_pages_per_second_upper_estimate": 0.004039237516055406
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19376.828192449066,
-      "required_pages_per_second_upper_estimate": 0.004035682311164355
+      "required_token_positions_per_second": 19295.146789095517,
+      "required_pages_per_second_upper_estimate": 0.004018670228929273
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29065.2422886736,
-      "required_pages_per_second_upper_estimate": 0.001742416895150696
+      "required_token_positions_per_second": 28942.720183643276,
+      "required_pages_per_second_upper_estimate": 0.0017350718820345501
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29065.2422886736,
-      "required_pages_per_second_upper_estimate": 0.0022596947434545666
+      "required_token_positions_per_second": 28942.720183643276,
+      "required_pages_per_second_upper_estimate": 0.0022501691887062444
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29065.2422886736,
-      "required_pages_per_second_upper_estimate": 0.37281772025337795
+      "required_token_positions_per_second": 28942.720183643276,
+      "required_pages_per_second_upper_estimate": 0.3712461382440358
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09447067775784035,
-    "producer_p95_seconds": 0.2864242590003414,
-    "gpu_update_seconds": 0.9512392749182111,
-    "producer_batches_per_second": 10.585295074979047,
-    "gpu_batches_per_second": 1.0512602100938078,
-    "producer_headroom": 10.06914841191838,
+    "producer_batch_seconds": 0.130393235007773,
+    "producer_p95_seconds": 0.3090809419991274,
+    "gpu_update_seconds": 0.9552661195828105,
+    "producer_batches_per_second": 7.669109520447039,
+    "gpu_batches_per_second": 1.046828710345894,
+    "producer_headroom": 7.326040492253031,
     "sustainable": true,
-    "buffer_seconds": 60.87931359476551,
+    "buffer_seconds": 60.18176553371706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273474.57808687724,
-    "upload_bytes_per_second": 5515711.498116179,
+    "generation_bytes_per_second": 263618.844054012,
+    "upload_bytes_per_second": 5538223.543179017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.68127829508531,
+    "estimated_full_shards_per_hour": 56.56646720137853,
     "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": 1.143838866999431,
-      "bytes": 6205051,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 3.739576465002756,
       "bytes": 2400719,
@@ -562,6 +556,12 @@
       "bytes": 11047745,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.0850444019997667,
+      "bytes": 7634525,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -311,3 +311,4 @@
 {"step": 471300, "loss": 0.5933566689491272, "lr": 0.0003, "grad_norm": 0.15876005156108036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68983851162, "tokens_per_second": 150041.98257796268, "data_wait_s": 1.570899985381402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 471400, "loss": 0.5182668566703796, "lr": 0.0003, "grad_norm": 0.15452290952364842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68998336149, "tokens_per_second": 96471.94536990493, "data_wait_s": 1.16259980131872e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 471500, "loss": 0.5422242879867554, "lr": 0.0003, "grad_norm": 0.16872998215867072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69012860781, "tokens_per_second": 150883.2237521362, "data_wait_s": 1.3167998986318707e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 471600, "loss": 0.5859622359275818, "lr": 0.0003, "grad_norm": 0.1678073753660921, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69027484151, "tokens_per_second": 152556.14185780848, "data_wait_s": 1.3237000530352816e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0489ce049b888937982f69dc56bd65e91ad4b1b4e954af111b443a640c0db383
-size 667797
+oid sha256:9803e4a568cdc4c498cbafb76f264cbb5c2ab6832f841a86f040c58aeb67c6fe
+size 669915