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": 75446.6721857848,
-      "required_pages_per_second_upper_estimate": 0.005219575842210679
+      "required_token_positions_per_second": 75816.25664033982,
+      "required_pages_per_second_upper_estimate": 0.005245144552330878
     },
     {
       "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": 9430.8340232231,
-      "required_pages_per_second_upper_estimate": 0.003948493267314517
+      "required_token_positions_per_second": 9477.032080042478,
+      "required_pages_per_second_upper_estimate": 0.003967835428979661
     },
     {
       "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": 18861.6680464462,
-      "required_pages_per_second_upper_estimate": 0.003928388038438606
+      "required_token_positions_per_second": 18954.064160084956,
+      "required_pages_per_second_upper_estimate": 0.003947631712260185
     },
     {
       "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": 28292.5020696693,
-      "required_pages_per_second_upper_estimate": 0.0016960922989273839
+      "required_token_positions_per_second": 28431.096240127434,
+      "required_pages_per_second_upper_estimate": 0.001704400807825304
     },
     {
       "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": 28292.5020696693,
-      "required_pages_per_second_upper_estimate": 0.002199617590351996
+      "required_token_positions_per_second": 28431.096240127434,
+      "required_pages_per_second_upper_estimate": 0.002210392677493786
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28292.5020696693,
-      "required_pages_per_second_upper_estimate": 0.36290583842779467
+      "required_token_positions_per_second": 28431.096240127434,
+      "required_pages_per_second_upper_estimate": 0.36468357563560544
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08674492598419192,
-    "producer_p95_seconds": 0.23453101300037815,
-    "gpu_update_seconds": 0.977220039850762,
-    "producer_batches_per_second": 11.52805179846757,
-    "gpu_batches_per_second": 1.0233109834226455,
-    "producer_headroom": 11.265443237900127,
+    "producer_batch_seconds": 0.06459825110940187,
+    "producer_p95_seconds": 0.21487318200161099,
+    "gpu_update_seconds": 0.972456347320784,
+    "producer_batches_per_second": 15.480295252984895,
+    "gpu_batches_per_second": 1.028323793407387,
+    "producer_headroom": 15.053911377164964,
     "sustainable": true,
-    "buffer_seconds": 60.58764247074724,
+    "buffer_seconds": 60.29229353388861,
     "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": 298503.88936027576,
-    "upload_bytes_per_second": 3854490.630778565,
+    "generation_bytes_per_second": 316494.4040166926,
+    "upload_bytes_per_second": 3844140.646809693,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.05198584181026,
+    "estimated_full_shards_per_hour": 67.91233148933013,
     "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": 1.62565908400029,
-      "bytes": 7724112,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.4196104830007243,
       "bytes": 988295,
@@ -562,6 +556,12 @@
       "bytes": 7551970,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.5439765690025524,
+      "bytes": 6266973,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -185,3 +185,4 @@
 {"step": 818500, "loss": 0.4756673276424408, "lr": 3e-05, "grad_norm": 0.27991657973072104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119805262667, "tokens_per_second": 151835.13705665397, "data_wait_s": 6.9779998739250004e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 818600, "loss": 0.5501465797424316, "lr": 3e-05, "grad_norm": 0.28155200234403827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119819942529, "tokens_per_second": 151043.82768929985, "data_wait_s": 9.424999007023871e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 818700, "loss": 0.565858781337738, "lr": 3e-05, "grad_norm": 0.27960762824885566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119834757197, "tokens_per_second": 153248.86429755297, "data_wait_s": 8.236998837674037e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 818800, "loss": 0.5714428424835205, "lr": 3e-05, "grad_norm": 0.2975220904316392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119849479219, "tokens_per_second": 152612.43304282203, "data_wait_s": 1.0782998288050294e-05, "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:afbe85f4058009e3b282736faf57e04141cfbe03ef59f98204f8c858d0dc9ca3
-size 398580
+oid sha256:fe079d66ac7bdb9a0ced224d9fe274c1b44803dc8ffc6b0676e1c8f25f792864
+size 400698