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": 76952.79484832539,
-      "required_pages_per_second_upper_estimate": 0.005323772902691288
+      "required_token_positions_per_second": 76614.55693402271,
+      "required_pages_per_second_upper_estimate": 0.0053003728189597405
     },
     {
       "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": 9619.099356040673,
-      "required_pages_per_second_upper_estimate": 0.004027316030738028
+      "required_token_positions_per_second": 9576.819616752839,
+      "required_pages_per_second_upper_estimate": 0.004009614386799566
     },
     {
       "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": 19238.198712081346,
-      "required_pages_per_second_upper_estimate": 0.0040068094463090054
+      "required_token_positions_per_second": 19153.639233505677,
+      "required_pages_per_second_upper_estimate": 0.003989197936905103
     },
     {
       "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": 28857.29806812202,
-      "required_pages_per_second_upper_estimate": 0.0017299509566410692
+      "required_token_positions_per_second": 28730.458850258517,
+      "required_pages_per_second_upper_estimate": 0.0017223471392024238
     },
     {
       "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": 28857.29806812202,
-      "required_pages_per_second_upper_estimate": 0.0022435279949566444
+      "required_token_positions_per_second": 28730.458850258517,
+      "required_pages_per_second_upper_estimate": 0.002233666803674515
     },
     {
       "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": 28857.29806812202,
-      "required_pages_per_second_upper_estimate": 0.37015043506525086
+      "required_token_positions_per_second": 28730.458850258517,
+      "required_pages_per_second_upper_estimate": 0.3685234777678393
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0932940619376268,
-    "producer_p95_seconds": 0.28653629300242756,
-    "gpu_update_seconds": 0.9580938566990128,
-    "producer_batches_per_second": 10.71879580791075,
-    "gpu_batches_per_second": 1.0437390794314967,
-    "producer_headroom": 10.269612414770423,
+    "producer_batch_seconds": 0.10472904607036071,
+    "producer_p95_seconds": 0.2718338570011838,
+    "gpu_update_seconds": 0.9623236490617769,
+    "producer_batches_per_second": 9.548449427565341,
+    "gpu_batches_per_second": 1.039151434109466,
+    "producer_headroom": 9.188698696016514,
     "sustainable": true,
-    "buffer_seconds": 60.359912972037804,
+    "buffer_seconds": 60.62638989089194,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317984.11004797113,
-    "upload_bytes_per_second": 3278919.477554647,
+    "generation_bytes_per_second": 319332.96668434027,
+    "upload_bytes_per_second": 3230278.0044667525,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.23198772506095,
+    "estimated_full_shards_per_hour": 68.5214209594503,
     "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": 1.9117556889978005,
-      "bytes": 10784078,
-      "pages": 43,
-      "shards": 3
-    },
     {
       "seconds": 1.6485553290003736,
       "bytes": 917525,
@@ -562,6 +556,12 @@
       "bytes": 6397585,
       "pages": 42,
       "shards": 2
+    },
+    {
+      "seconds": 1.4411041519997525,
+      "bytes": 2727332,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -287,3 +287,4 @@
 {"step": 828700, "loss": 0.5343151688575745, "lr": 3e-05, "grad_norm": 0.28000622198113484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121298329142, "tokens_per_second": 128172.62825832651, "data_wait_s": 2.0457999198697507e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 828800, "loss": 0.5465978980064392, "lr": 3e-05, "grad_norm": 0.2814293850838141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121313035457, "tokens_per_second": 152555.76097589397, "data_wait_s": 7.995000487426296e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 828900, "loss": 0.5660039782524109, "lr": 3e-05, "grad_norm": 0.29373261987504706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121327800196, "tokens_per_second": 152236.0218599373, "data_wait_s": 1.551900277263485e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 829000, "loss": 0.5336809158325195, "lr": 3e-05, "grad_norm": 0.3212830800290023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121342467616, "tokens_per_second": 151047.79262992184, "data_wait_s": 1.3477001630235463e-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 sha256:762f3787c6c1732c637e7bf36bd0c5fa61375836a92588004d517443b63f6659
-size 616182
+oid sha256:ee270621157d26971a124484bad757d8ac0ebb76f28f16866392959b4d323bf4
+size 618300