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": 75968.17661473945,
-      "required_pages_per_second_upper_estimate": 0.005255654728662751
+      "required_token_positions_per_second": 76128.5698897275,
+      "required_pages_per_second_upper_estimate": 0.005266751107590096
     },
     {
       "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": 9496.022076842431,
-      "required_pages_per_second_upper_estimate": 0.003975786144083575
+      "required_token_positions_per_second": 9516.071236215937,
+      "required_pages_per_second_upper_estimate": 0.003984180308439205
     },
     {
       "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": 18992.044153684863,
-      "required_pages_per_second_upper_estimate": 0.003955541943327259
+      "required_token_positions_per_second": 19032.142472431875,
+      "required_pages_per_second_upper_estimate": 0.003963893365658485
     },
     {
       "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": 28488.066230527296,
-      "required_pages_per_second_upper_estimate": 0.0017078160717616372
+      "required_token_positions_per_second": 28548.213708647814,
+      "required_pages_per_second_upper_estimate": 0.0017114218212350809
     },
     {
       "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": 28488.066230527296,
-      "required_pages_per_second_upper_estimate": 0.00221482184366287
+      "required_token_positions_per_second": 28548.213708647814,
+      "required_pages_per_second_upper_estimate": 0.002219498045525944
     },
     {
       "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": 28488.066230527296,
-      "required_pages_per_second_upper_estimate": 0.36541432550284497
+      "required_token_positions_per_second": 28548.213708647814,
+      "required_pages_per_second_upper_estimate": 0.3661858327708446
     }
   ],
   "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": 0.4674527591718771,
-    "producer_p95_seconds": 0.2626089660006983,
-    "gpu_update_seconds": 0.9705116442888954,
-    "producer_batches_per_second": 2.1392536045173096,
-    "gpu_batches_per_second": 1.0303843399351598,
-    "producer_headroom": 2.0761705332710405,
+    "producer_batch_seconds": 0.06187203408596531,
+    "producer_p95_seconds": 0.22614036400045734,
+    "gpu_update_seconds": 0.9684668989158113,
+    "producer_batches_per_second": 16.162390889082378,
+    "gpu_batches_per_second": 1.0325598129574585,
+    "producer_headroom": 15.652740583414772,
     "sustainable": true,
-    "buffer_seconds": 60.171721945911514,
+    "buffer_seconds": 60.0449477327803,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447368.30609686836,
-    "upload_bytes_per_second": 3888536.367683161,
+    "generation_bytes_per_second": 449376.9926912854,
+    "upload_bytes_per_second": 3924257.8442087774,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.99482428721942,
+    "estimated_full_shards_per_hour": 96.42584167055055,
     "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": 1.5443801999999778,
-      "bytes": 7496481,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.4706297409993567,
       "bytes": 5704544,
@@ -562,6 +556,12 @@
       "bytes": 5688723,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.593651556999248,
+      "bytes": 11614889,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -143,3 +143,4 @@
 {"step": 814300, "loss": 0.5602394938468933, "lr": 3e-05, "grad_norm": 0.27729984187245255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119190194816, "tokens_per_second": 151159.33371674424, "data_wait_s": 7.97600114310626e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 814400, "loss": 0.5305436253547668, "lr": 3e-05, "grad_norm": 0.27287402219806217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119204952198, "tokens_per_second": 129982.8005996017, "data_wait_s": 8.906999937607907e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 814500, "loss": 0.5320994257926941, "lr": 3e-05, "grad_norm": 0.2902150563963192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119219658989, "tokens_per_second": 152389.13778780532, "data_wait_s": 1.1037000149372034e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 814600, "loss": 0.5440930128097534, "lr": 3e-05, "grad_norm": 0.2911424202091485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119234400162, "tokens_per_second": 152597.91513367646, "data_wait_s": 1.3880000551580451e-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:a50364d5bd8d169449f8120ad8e9b6ffde82599a64e9063b2cbff8782ca49833
-size 308841
+oid sha256:11dea8acd75b7172a2f5734996591f3c56fdba207b53a6a5d77f06b7d608cf0c
+size 310959