Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75606.79534556897,
-      "required_pages_per_second_upper_estimate": 0.006553405396883024
+      "required_token_positions_per_second": 75609.18069986637,
+      "required_pages_per_second_upper_estimate": 0.006553612153347897
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9450.849418196121,
-      "required_pages_per_second_upper_estimate": 0.003770085135709319
+      "required_token_positions_per_second": 9451.147587483296,
+      "required_pages_per_second_upper_estimate": 0.0037702040798960014
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18901.698836392243,
-      "required_pages_per_second_upper_estimate": 0.004372575367261525
+      "required_token_positions_per_second": 18902.29517496659,
+      "required_pages_per_second_upper_estimate": 0.0043727133196954995
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28352.548254588368,
-      "required_pages_per_second_upper_estimate": 0.0017265214343800796
+      "required_token_positions_per_second": 28353.442762449886,
+      "required_pages_per_second_upper_estimate": 0.001726575905215726
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28352.548254588368,
-      "required_pages_per_second_upper_estimate": 0.002182657649515037
+      "required_token_positions_per_second": 28353.442762449886,
+      "required_pages_per_second_upper_estimate": 0.0021827265112063727
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28352.548254588368,
-      "required_pages_per_second_upper_estimate": 0.3520437594470662
+      "required_token_positions_per_second": 28353.442762449886,
+      "required_pages_per_second_upper_estimate": 0.352054866240981
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10743396863267662,
-    "producer_p95_seconds": 0.3628431940014707,
-    "gpu_update_seconds": 0.9751504433301036,
-    "producer_batches_per_second": 9.308043002851935,
-    "gpu_batches_per_second": 1.0254827927730168,
-    "producer_headroom": 9.076742260766734,
+    "producer_batch_seconds": 0.140550233101294,
+    "producer_p95_seconds": 0.37611892700078897,
+    "gpu_update_seconds": 0.9751196788213619,
+    "producer_batches_per_second": 7.11489392749213,
+    "gpu_batches_per_second": 1.025515146211295,
+    "producer_headroom": 6.937873081424184,
     "sustainable": true,
-    "buffer_seconds": 60.45932748646642,
+    "buffer_seconds": 60.45742008692444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526660.7471500915,
+    "generation_bytes_per_second": 507631.4843254223,
     "upload_bytes_per_second": 6214687.544045871,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.00913630368288,
+    "estimated_full_shards_per_hour": 108.92589948007586,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -324,3 +324,4 @@
 {"step": 32400, "loss": 0.6780863795429468, "lr": 0.0003, "grad_norm": 0.12556539759495675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4741489249, "tokens_per_second": 147126.84428959162, "data_wait_s": 9.2410045908764e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 32500, "loss": 0.6430358495563269, "lr": 0.0003, "grad_norm": 0.13856002775264428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4756103473, "tokens_per_second": 149169.28847649784, "data_wait_s": 1.808099477784708e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32600, "loss": 0.6330348439514637, "lr": 0.0003, "grad_norm": 0.13050141602012488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4770742301, "tokens_per_second": 149874.27247702537, "data_wait_s": 9.467003110330552e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 32700, "loss": 0.6028679460287094, "lr": 0.0003, "grad_norm": 0.13054031573608635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4785354945, "tokens_per_second": 148592.2659081701, "data_wait_s": 1.6526995750609785e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47c7b991e2763d72398ad1582dd0ef7f3699f36debc14a340b7cf2e772dbd693
-size 689557
+oid sha256:4133881df98b7b8cb589e9c5711070ba43dcf0537caefd77cb5d0b5e8d98ff40
+size 691675