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": 75560.36214649593,
-      "required_pages_per_second_upper_estimate": 0.005227441177489728
+      "required_token_positions_per_second": 76634.53374917958,
+      "required_pages_per_second_upper_estimate": 0.00530175486138489
     },
     {
       "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": 9445.045268311991,
-      "required_pages_per_second_upper_estimate": 0.003954443218868692
+      "required_token_positions_per_second": 9579.316718647447,
+      "required_pages_per_second_upper_estimate": 0.004010659871217422
     },
     {
       "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": 18890.090536623982,
-      "required_pages_per_second_upper_estimate": 0.003934307693591167
+      "required_token_positions_per_second": 19158.633437294895,
+      "required_pages_per_second_upper_estimate": 0.003990238097848387
     },
     {
       "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": 28335.13580493597,
-      "required_pages_per_second_upper_estimate": 0.0016986481262586763
+      "required_token_positions_per_second": 28737.950155942344,
+      "required_pages_per_second_upper_estimate": 0.0017227962315396078
     },
     {
       "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": 28335.13580493597,
-      "required_pages_per_second_upper_estimate": 0.002202932175742997
+      "required_token_positions_per_second": 28737.950155942344,
+      "required_pages_per_second_upper_estimate": 0.002234249219740661
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28335.13580493597,
-      "required_pages_per_second_upper_estimate": 0.36345269820725706
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28737.950155942344,
+      "required_pages_per_second_upper_estimate": 0.368619568193614
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.39806838436715,
-    "producer_p95_seconds": 0.24290076200122712,
-    "gpu_update_seconds": 0.975749690784391,
-    "producer_batches_per_second": 2.512131179645935,
-    "gpu_batches_per_second": 1.0248530022039921,
-    "producer_headroom": 2.4512112217493485,
+    "producer_batch_seconds": 0.08469534066421147,
+    "producer_p95_seconds": 0.23999202900085947,
+    "gpu_update_seconds": 0.9620727939874665,
+    "producer_batches_per_second": 11.80702494561848,
+    "gpu_batches_per_second": 1.0394223870060166,
+    "producer_headroom": 11.359217478110885,
     "sustainable": true,
-    "buffer_seconds": 60.496480828632244,
+    "buffer_seconds": 60.61058602121039,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256397.28062738074,
-    "upload_bytes_per_second": 3905020.038219226,
+    "generation_bytes_per_second": 268074.83698201075,
+    "upload_bytes_per_second": 4001900.3118129503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.016887799416224,
+    "estimated_full_shards_per_hour": 57.5226195535206,
     "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": 1.720672621000631,
-      "bytes": 611156,
-      "pages": 21,
-      "shards": 1
-    },
     {
       "seconds": 1.6194256800008588,
       "bytes": 9820718,
@@ -562,6 +556,12 @@
       "bytes": 2327821,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.6191135740009486,
+      "bytes": 10591273,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -192,3 +192,4 @@
 {"step": 819200, "loss": 0.5278240442276001, "lr": 3e-05, "grad_norm": 0.28942664777665683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119907919956, "tokens_per_second": 152879.20086366482, "data_wait_s": 1.1234998964937404e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 819300, "loss": 0.5462301969528198, "lr": 3e-05, "grad_norm": 0.2886934143894675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119922553260, "tokens_per_second": 151621.18268768938, "data_wait_s": 9.847000910667703e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 819400, "loss": 0.5596259832382202, "lr": 3e-05, "grad_norm": 0.2890894522266533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119937252394, "tokens_per_second": 152099.39793271886, "data_wait_s": 1.1822998203570023e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 819500, "loss": 0.5511067509651184, "lr": 3e-05, "grad_norm": 0.2806126738103894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119951999815, "tokens_per_second": 152030.24434130485, "data_wait_s": 9.433999366592616e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6dd6edcf65e23a93c780e28003c92518561fe2a4b12ad71b5d678337fc2bd511
-size 413406
+oid sha256:fb9916d16f33336be7774d701c8bd18c059da220ebb54537686f3c37b1e1e7b2
+size 415524