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": 77243.65984217785,
-      "required_pages_per_second_upper_estimate": 0.00534389561786577
+      "required_token_positions_per_second": 77329.65729477121,
+      "required_pages_per_second_upper_estimate": 0.00534984512117776
     },
     {
       "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": 9655.45748027223,
-      "required_pages_per_second_upper_estimate": 0.004042538418109811
+      "required_token_positions_per_second": 9666.207161846402,
+      "required_pages_per_second_upper_estimate": 0.004047039085306038
     },
     {
       "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": 19310.91496054446,
-      "required_pages_per_second_upper_estimate": 0.004021954323207444
+      "required_token_positions_per_second": 19332.414323692803,
+      "required_pages_per_second_upper_estimate": 0.004026432073574909
     },
     {
       "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": 28966.372440816696,
-      "required_pages_per_second_upper_estimate": 0.001736489798737191
+      "required_token_positions_per_second": 28998.621485539206,
+      "required_pages_per_second_upper_estimate": 0.0017384230797268655
     },
     {
       "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": 28966.372440816696,
-      "required_pages_per_second_upper_estimate": 0.0022520080476661933
+      "required_token_positions_per_second": 28998.621485539206,
+      "required_pages_per_second_upper_estimate": 0.002254515269044812
     },
     {
       "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": 28966.372440816696,
-      "required_pages_per_second_upper_estimate": 0.3715495240032413
+      "required_token_positions_per_second": 28998.621485539206,
+      "required_pages_per_second_upper_estimate": 0.37196318012261526
     }
   ],
   "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": 0.10673617788302181,
-    "producer_p95_seconds": 0.30406462300015846,
-    "gpu_update_seconds": 0.9544861047578409,
-    "producer_batches_per_second": 9.368894594445347,
-    "gpu_batches_per_second": 1.0476841883975947,
-    "producer_headroom": 8.942479707338931,
+    "producer_batch_seconds": 0.08182209437489973,
+    "producer_p95_seconds": 0.29354966200116905,
+    "gpu_update_seconds": 0.9534246313669523,
+    "producer_batches_per_second": 12.221637781821023,
+    "gpu_batches_per_second": 1.0488506034989584,
+    "producer_headroom": 11.652410496833125,
     "sustainable": true,
-    "buffer_seconds": 60.13262459974398,
+    "buffer_seconds": 60.065751776118,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 499409.0941177699,
-    "upload_bytes_per_second": 4853999.2363889,
+    "generation_bytes_per_second": 501324.4293938101,
+    "upload_bytes_per_second": 4920855.024541673,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.16156594896148,
+    "estimated_full_shards_per_hour": 107.57255231247642,
     "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": 1.2272778379992815,
-      "bytes": 1850276,
-      "pages": 12,
-      "shards": 2
-    },
     {
       "seconds": 1.1599697590027063,
       "bytes": 11845234,
@@ -562,6 +556,12 @@
       "bytes": 6220019,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9491098379985488,
+      "bytes": 6427108,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 792900, "loss": 0.5092544555664062, "lr": 3e-05, "grad_norm": 0.2613729770209938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116058026192, "tokens_per_second": 151748.72570455764, "data_wait_s": 1.908899866975844e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 793000, "loss": 0.5120710134506226, "lr": 3e-05, "grad_norm": 0.2728636435751163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116072586789, "tokens_per_second": 151803.31569046908, "data_wait_s": 1.4789002307225019e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 793100, "loss": 0.4716569483280182, "lr": 3e-05, "grad_norm": 0.2565845044052931, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116087179076, "tokens_per_second": 152250.3484565741, "data_wait_s": 1.3458000466926023e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 793200, "loss": 0.5424240231513977, "lr": 3e-05, "grad_norm": 0.25608720528697326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116101808494, "tokens_per_second": 151502.9017143151, "data_wait_s": 9.085000783670694e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:abad4ef0c0c534340e13e5b93ecce80f51c54c8e2430e008bcc734be78c94f8e
-size 705921
+oid sha256:e246b944f00698ebf31ff0d0e9051bfd16594a2f240fe36609a42b1f9947f8fa
+size 708039