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": 75570.2741855813,
-      "required_pages_per_second_upper_estimate": 0.0065502398352973255
+      "required_token_positions_per_second": 75536.72656155463,
+      "required_pages_per_second_upper_estimate": 0.006547332012272371
     },
     {
       "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": 9446.284273197662,
-      "required_pages_per_second_upper_estimate": 0.0037682640311144334
+      "required_token_positions_per_second": 9442.09082019433,
+      "required_pages_per_second_upper_estimate": 0.0037665912000136945
     },
     {
       "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": 18892.568546395323,
-      "required_pages_per_second_upper_estimate": 0.004370463235358353
+      "required_token_positions_per_second": 18884.18164038866,
+      "required_pages_per_second_upper_estimate": 0.004368523072258216
     },
     {
       "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": 28338.852819592983,
-      "required_pages_per_second_upper_estimate": 0.001725687454243255
+      "required_token_positions_per_second": 28326.27246058299,
+      "required_pages_per_second_upper_estimate": 0.0017249213763835891
     },
     {
       "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": 28338.852819592983,
-      "required_pages_per_second_upper_estimate": 0.0021816033370177018
+      "required_token_positions_per_second": 28326.27246058299,
+      "required_pages_per_second_upper_estimate": 0.00218063486615645
     },
     {
       "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": 28338.852819592983,
-      "required_pages_per_second_upper_estimate": 0.3518737079800959
+      "required_token_positions_per_second": 28326.27246058299,
+      "required_pages_per_second_upper_estimate": 0.35171750202494495
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.434673400016095,
-    "producer_p95_seconds": 0.3332169179993798,
-    "gpu_update_seconds": 0.9756217083312794,
-    "producer_batches_per_second": 2.300577859061475,
-    "gpu_batches_per_second": 1.024987442838288,
-    "producer_headroom": 2.2444937010066734,
+    "producer_batch_seconds": 0.4408702201640722,
+    "producer_p95_seconds": 0.33095603300171206,
+    "gpu_update_seconds": 0.9760550047124333,
+    "producer_batches_per_second": 2.268241206284799,
+    "gpu_batches_per_second": 1.024532424066225,
+    "producer_headroom": 2.213928181289245,
     "sustainable": true,
-    "buffer_seconds": 60.48854591653932,
+    "buffer_seconds": 60.51541029217087,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289673.2651531505,
-    "upload_bytes_per_second": 5567270.98190291,
+    "generation_bytes_per_second": 283101.0000492304,
+    "upload_bytes_per_second": 5601371.258226041,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.15713945337187,
+    "estimated_full_shards_per_hour": 60.746884356572004,
     "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": 0.8255374049986131,
-      "bytes": 1098294,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 0.9614813039988803,
       "bytes": 5875069,
@@ -562,6 +556,12 @@
       "bytes": 7732628,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.990468955002143,
+      "bytes": 4589915,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -251,3 +251,4 @@
 {"step": 305300, "loss": 0.592576090246439, "lr": 0.0003, "grad_norm": 0.16281926172888336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44687979341, "tokens_per_second": 151004.13191365328, "data_wait_s": 9.408999176230282e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 305400, "loss": 0.5878156870603561, "lr": 0.0003, "grad_norm": 0.14787342915316243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44702672078, "tokens_per_second": 151011.75474795562, "data_wait_s": 9.897994459606707e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 305500, "loss": 0.584441065788269, "lr": 0.0003, "grad_norm": 0.18012175779245443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44717283910, "tokens_per_second": 139829.5428343176, "data_wait_s": 1.553699985379353e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 305600, "loss": 0.5479111261665821, "lr": 0.0003, "grad_norm": 0.1525635898032803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44731776385, "tokens_per_second": 138816.0164328992, "data_wait_s": 1.2979995517525822e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff6a8b1c0ff8fd3b03575a4992bbc4a9014470c6e2b2c3dd5ecb569f2cd28b04
-size 539934
+oid sha256:3d6710234ccc8b646969138476e5f2d65535bbe794652457654e7d6deef51fca
+size 542052