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": 75699.36746562648,
-      "required_pages_per_second_upper_estimate": 0.0065614293136277155
+      "required_token_positions_per_second": 75655.81659982461,
+      "required_pages_per_second_upper_estimate": 0.006557654434958669
     },
     {
       "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": 9462.42093320331,
-      "required_pages_per_second_upper_estimate": 0.0037747011860552538
+      "required_token_positions_per_second": 9456.977074978076,
+      "required_pages_per_second_upper_estimate": 0.003772529549616274
     },
     {
       "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": 18924.84186640662,
-      "required_pages_per_second_upper_estimate": 0.004377929099951934
+      "required_token_positions_per_second": 18913.954149956153,
+      "required_pages_per_second_upper_estimate": 0.004375410418368383
     },
     {
       "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": 28387.26279960993,
-      "required_pages_per_second_upper_estimate": 0.001728635368038751
+      "required_token_positions_per_second": 28370.93122493423,
+      "required_pages_per_second_upper_estimate": 0.0017276408608261512
     },
     {
       "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": 28387.26279960993,
-      "required_pages_per_second_upper_estimate": 0.0021853300712868077
+      "required_token_positions_per_second": 28370.93122493423,
+      "required_pages_per_second_upper_estimate": 0.00218407282145958
     },
     {
       "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": 28387.26279960993,
-      "required_pages_per_second_upper_estimate": 0.35247479791412556
+      "required_token_positions_per_second": 28370.93122493423,
+      "required_pages_per_second_upper_estimate": 0.35227201441491773
     }
   ],
   "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": 0.10502399002331231,
-    "producer_p95_seconds": 0.33523229300044477,
-    "gpu_update_seconds": 0.9739579400511948,
-    "producer_batches_per_second": 9.521634055019513,
-    "gpu_batches_per_second": 1.0267383825090397,
-    "producer_headroom": 9.27367109014811,
+    "producer_batch_seconds": 0.08877195110164848,
+    "producer_p95_seconds": 0.3551546849994338,
+    "gpu_update_seconds": 0.9745185937252961,
+    "producer_batches_per_second": 11.264819434406125,
+    "gpu_batches_per_second": 1.0261476860870309,
+    "producer_headroom": 10.977775993786842,
     "sustainable": true,
-    "buffer_seconds": 60.38539228317408,
+    "buffer_seconds": 60.42015281096836,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299869.45925812615,
-    "upload_bytes_per_second": 5754785.797824073,
+    "generation_bytes_per_second": 298717.96207559755,
+    "upload_bytes_per_second": 5676065.677209759,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.34500535304869,
+    "estimated_full_shards_per_hour": 64.09792086316057,
     "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": 1.1060321420009132,
-      "bytes": 6015087,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.2197453389999282,
       "bytes": 6685693,
@@ -562,6 +556,12 @@
       "bytes": 5988067,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.234346370001731,
+      "bytes": 974463,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -287,3 +287,4 @@
 {"step": 308900, "loss": 0.6228929907083511, "lr": 0.0003, "grad_norm": 0.16690376191349374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45214920191, "tokens_per_second": 150615.64277173474, "data_wait_s": 1.103100657928735e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 309000, "loss": 0.601045886054635, "lr": 0.0003, "grad_norm": 0.1612058812513313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45229451852, "tokens_per_second": 149826.7964951862, "data_wait_s": 9.901996236294508e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 309100, "loss": 0.5872989073395729, "lr": 0.0003, "grad_norm": 0.15178216332064237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45243936418, "tokens_per_second": 148635.3064319802, "data_wait_s": 1.1945994629058987e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 309200, "loss": 0.5954781509935856, "lr": 0.0003, "grad_norm": 0.13767809501678832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45258524838, "tokens_per_second": 149608.27499610386, "data_wait_s": 1.2308002624195069e-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 sha256:26c6916926872e7adcc288a47182f7374ecf2ba3adc998500ce6f11887d7593a
-size 616182
+oid sha256:aee435b9b509ffa91131a46e9be700e46b2b619707409277211e984803bb9a07
+size 618300