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": 76382.47571916689,
-      "required_pages_per_second_upper_estimate": 0.005284316902013406
+      "required_token_positions_per_second": 75661.34979537701,
+      "required_pages_per_second_upper_estimate": 0.005234427737362951
     },
     {
       "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": 9547.809464895861,
-      "required_pages_per_second_upper_estimate": 0.003997468442017908
+      "required_token_positions_per_second": 9457.668724422127,
+      "required_pages_per_second_upper_estimate": 0.003959728396334266
     },
     {
       "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": 19095.618929791723,
-      "required_pages_per_second_upper_estimate": 0.00397711383773979
+      "required_token_positions_per_second": 18915.337448844253,
+      "required_pages_per_second_upper_estimate": 0.003939565959600801
     },
     {
       "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": 28643.428394687584,
-      "required_pages_per_second_upper_estimate": 0.0017171297962787575
+      "required_token_positions_per_second": 28373.006173266378,
+      "required_pages_per_second_upper_estimate": 0.00170091839701538
     },
     {
       "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": 28643.428394687584,
-      "required_pages_per_second_upper_estimate": 0.0022269005685603927
+      "required_token_positions_per_second": 28373.006173266378,
+      "required_pages_per_second_upper_estimate": 0.00220587643030654
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28643.428394687584,
-      "required_pages_per_second_upper_estimate": 0.367407144529798
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28373.006173266378,
+      "required_pages_per_second_upper_estimate": 0.3639384586301661
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10931410808596809,
-    "producer_p95_seconds": 0.2239400950002164,
-    "gpu_update_seconds": 0.9652475820641567,
-    "producer_batches_per_second": 9.147950045144842,
-    "gpu_batches_per_second": 1.036003631173596,
-    "producer_headroom": 8.830036661919753,
+    "producer_batch_seconds": 0.35371063149210613,
+    "producer_p95_seconds": 0.26256509400081995,
+    "gpu_update_seconds": 0.9744473261367175,
+    "producer_batches_per_second": 2.8271697567629297,
+    "gpu_batches_per_second": 1.0262227348548314,
+    "producer_headroom": 2.754928010012231,
     "sustainable": true,
-    "buffer_seconds": 60.81059767004187,
+    "buffer_seconds": 60.41573422047648,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 519157.75506667583,
+    "generation_bytes_per_second": 450321.8281006348,
     "upload_bytes_per_second": 5129754.504081647,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.39916886329847,
+    "estimated_full_shards_per_hour": 96.62858135475429,
     "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": [

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

@@ -158,3 +158,4 @@
 {"step": 775800, "loss": 0.5179250240325928, "lr": 3e-05, "grad_norm": 0.2545529767185388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113554958333, "tokens_per_second": 150953.8647159574, "data_wait_s": 1.4745999578735791e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775900, "loss": 0.520417332649231, "lr": 3e-05, "grad_norm": 0.2541022054553516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113569501543, "tokens_per_second": 151064.36756849402, "data_wait_s": 1.3685999874724075e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 776000, "loss": 0.53727126121521, "lr": 3e-05, "grad_norm": 0.2480641214078739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113584171181, "tokens_per_second": 151938.30523470917, "data_wait_s": 1.093399987439625e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 776100, "loss": 0.5250530242919922, "lr": 3e-05, "grad_norm": 0.2389914929713818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113598928558, "tokens_per_second": 152751.50531584598, "data_wait_s": 1.7178001144202426e-05, "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 sha256:91a90537b75f286c113af230dee12c6f01adb51e69036010b2f9e1a552d1607c
-size 341394
+oid sha256:740a81c5f8b1f1984083439d1039a99a08884ab6c5971921d39e77b9150d321c
+size 343512