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": 75459.14757308971,
-      "required_pages_per_second_upper_estimate": 0.006540607662174379
+      "required_token_positions_per_second": 75552.771757313,
+      "required_pages_per_second_upper_estimate": 0.0065487227691745295
     },
     {
       "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": 9432.393446636213,
-      "required_pages_per_second_upper_estimate": 0.0037627227727127068
+      "required_token_positions_per_second": 9444.096469664126,
+      "required_pages_per_second_upper_estimate": 0.0037673912835743282
     },
     {
       "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": 18864.786893272427,
-      "required_pages_per_second_upper_estimate": 0.004364036438848766
+      "required_token_positions_per_second": 18888.19293932825,
+      "required_pages_per_second_upper_estimate": 0.004369451015671441
     },
     {
       "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": 28297.180339908642,
-      "required_pages_per_second_upper_estimate": 0.0017231498188690823
+      "required_token_positions_per_second": 28332.28940899238,
+      "required_pages_per_second_upper_estimate": 0.0017252877769732323
     },
     {
       "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": 28297.180339908642,
-      "required_pages_per_second_upper_estimate": 0.002178395274174794
+      "required_token_positions_per_second": 28332.28940899238,
+      "required_pages_per_second_upper_estimate": 0.0021810980675009838
     },
     {
       "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": 28297.180339908642,
-      "required_pages_per_second_upper_estimate": 0.35135627525123414
+      "required_token_positions_per_second": 28332.28940899238,
+      "required_pages_per_second_upper_estimate": 0.3517922123867586
     }
   ],
   "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": 0.1160273873593951,
-    "producer_p95_seconds": 0.3607228969995049,
-    "gpu_update_seconds": 0.9770584796043061,
-    "producer_batches_per_second": 8.618654808648735,
-    "gpu_batches_per_second": 1.0234801916922975,
-    "producer_headroom": 8.420929763572675,
+    "producer_batch_seconds": 0.22890363539852387,
+    "producer_p95_seconds": 0.35882261200094945,
+    "gpu_update_seconds": 0.9758477192183702,
+    "producer_batches_per_second": 4.368650581975198,
+    "gpu_batches_per_second": 1.0247500509618193,
+    "producer_headroom": 4.263137706482503,
     "sustainable": true,
-    "buffer_seconds": 60.57762573546698,
+    "buffer_seconds": 60.50255859153895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 174,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 610213.002043595,
+    "generation_bytes_per_second": 577571.3660680001,
     "upload_bytes_per_second": 5146839.074837373,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.46875260343974,
+    "estimated_full_shards_per_hour": 61.96668499245644,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 145900, "loss": 0.6064195018261671, "lr": 0.0003, "grad_norm": 0.12443164934563086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21351993301, "tokens_per_second": 150695.9610669303, "data_wait_s": 1.3901000784244388e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146000, "loss": 0.6049474477767944, "lr": 0.0003, "grad_norm": 0.14668302956766735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21366663299, "tokens_per_second": 149947.5760015606, "data_wait_s": 1.5970999811543152e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146100, "loss": 0.6441834196448326, "lr": 0.0003, "grad_norm": 0.14358625311955592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21381439536, "tokens_per_second": 150884.7213961859, "data_wait_s": 8.031001925701275e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 146200, "loss": 0.616436455398798, "lr": 0.0003, "grad_norm": 0.1636305480061172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21396067541, "tokens_per_second": 150013.2628389867, "data_wait_s": 1.0538999049458653e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0000ddc26eceaa044684d430c4d1e90f1d30bb131287a13cf38b004f8be7cb9f
-size 663561
+oid sha256:37aed4d02ea87458963acfdabfa70c0c6f4ac1cfc298e20e814c0625fc909e7a
+size 665679