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": 75689.17022486347,
-      "required_pages_per_second_upper_estimate": 0.006560545442643048
+      "required_token_positions_per_second": 75942.61233808876,
+      "required_pages_per_second_upper_estimate": 0.0065825131626742524
     },
     {
       "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": 9461.146278107934,
-      "required_pages_per_second_upper_estimate": 0.003774192707079916
+      "required_token_positions_per_second": 9492.826542261095,
+      "required_pages_per_second_upper_estimate": 0.003786830438112771
     },
     {
       "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": 18922.292556215867,
-      "required_pages_per_second_upper_estimate": 0.00437733936190563
+      "required_token_positions_per_second": 18985.65308452219,
+      "required_pages_per_second_upper_estimate": 0.004391996705022085
     },
     {
       "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": 28383.4388343238,
-      "required_pages_per_second_upper_estimate": 0.001728402508615621
+      "required_token_positions_per_second": 28478.479626783283,
+      "required_pages_per_second_upper_estimate": 0.0017341899942359025
     },
     {
       "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": 28383.4388343238,
-      "required_pages_per_second_upper_estimate": 0.0021850356918536678
+      "required_token_positions_per_second": 28478.479626783283,
+      "required_pages_per_second_upper_estimate": 0.002192352195146951
     },
     {
       "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": 28383.4388343238,
-      "required_pages_per_second_upper_estimate": 0.35242731706326036
+      "required_token_positions_per_second": 28478.479626783283,
+      "required_pages_per_second_upper_estimate": 0.35360740562453635
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.6830518919609005,
-    "producer_p95_seconds": 0.21424734500033082,
-    "gpu_update_seconds": 0.9740891567573398,
-    "producer_batches_per_second": 1.4640176123796496,
-    "gpu_batches_per_second": 1.0266000735794198,
-    "producer_headroom": 1.4260836815207867,
+    "producer_batch_seconds": 0.0798477174297858,
+    "producer_p95_seconds": 0.20272005100014212,
+    "gpu_update_seconds": 0.9708383439823017,
+    "producer_batches_per_second": 12.523839530908962,
+    "gpu_batches_per_second": 1.0300376022418722,
+    "producer_headroom": 12.158623630487744,
     "sustainable": true,
-    "buffer_seconds": 60.39352771895506,
+    "buffer_seconds": 60.19197732690271,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311647.01342485653,
-    "upload_bytes_per_second": 5713807.3478162605,
+    "generation_bytes_per_second": 323091.6482308357,
+    "upload_bytes_per_second": 5757418.377995114,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.87219430979988,
+    "estimated_full_shards_per_hour": 69.32794652169993,
     "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": [
@@ -538,6 +538,18 @@
       "bytes": 2261871,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 2.0209359849995963,
+      "bytes": 15406607,
+      "pages": 37,
+      "shards": 4
+    },
+    {
+      "seconds": 1.1222496679984033,
+      "bytes": 5615350,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -98,3 +98,4 @@
 {"step": 364800, "loss": 0.5859208181500435, "lr": 0.0003, "grad_norm": 0.15828551047307055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53396639105, "tokens_per_second": 150056.78764171584, "data_wait_s": 7.472999641322531e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 364900, "loss": 0.5903571639209986, "lr": 0.0003, "grad_norm": 0.1696520185735211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53411391267, "tokens_per_second": 150548.5994235989, "data_wait_s": 1.2202001016703434e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365000, "loss": 0.6108045075088739, "lr": 0.0003, "grad_norm": 0.19939552588878048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53425992443, "tokens_per_second": 148894.1315876089, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 365100, "loss": 0.5924281161278486, "lr": 0.0003, "grad_norm": 0.14468772941650945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53440489453, "tokens_per_second": 127387.38159713705, "data_wait_s": 1.4104000001680106e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:feed769ef4e12f1dc1d41e6cc148a36d3917fd56bc3917d1876f3e33bc223f33
-size 213531
+oid sha256:46153bd74682c796543a4ae5caf097cc8a60ff912cd5ae5018e117c0b370bd0f
+size 215649