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": 75904.5897174523,
-      "required_pages_per_second_upper_estimate": 0.006579217458284935
+      "required_token_positions_per_second": 75712.83710133901,
+      "required_pages_per_second_upper_estimate": 0.006562596827512798
     },
     {
       "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": 9488.073714681537,
-      "required_pages_per_second_upper_estimate": 0.0037849344641301807
+      "required_token_positions_per_second": 9464.104637667377,
+      "required_pages_per_second_upper_estimate": 0.003775372840939595
     },
     {
       "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": 18976.147429363074,
-      "required_pages_per_second_upper_estimate": 0.004389797738994841
+      "required_token_positions_per_second": 18928.209275334753,
+      "required_pages_per_second_upper_estimate": 0.004378708090743084
     },
     {
       "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": 28464.22114404461,
-      "required_pages_per_second_upper_estimate": 0.001733321727445595
+      "required_token_positions_per_second": 28392.31391300213,
+      "required_pages_per_second_upper_estimate": 0.0017289429543431922
     },
     {
       "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": 28464.22114404461,
-      "required_pages_per_second_upper_estimate": 0.0021912545376757218
+      "required_token_positions_per_second": 28392.31391300213,
+      "required_pages_per_second_upper_estimate": 0.002185718919978116
     },
     {
       "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": 28464.22114404461,
-      "required_pages_per_second_upper_estimate": 0.35343036298899405
+      "required_token_positions_per_second": 28392.31391300213,
+      "required_pages_per_second_upper_estimate": 0.3525375158374676
     }
   ],
   "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": 0.2641105627733964,
-    "producer_p95_seconds": 0.19650123199971858,
-    "gpu_update_seconds": 0.9713246626382614,
-    "producer_batches_per_second": 3.7862930944491895,
-    "gpu_batches_per_second": 1.029521887443743,
-    "producer_headroom": 3.6777198626154384,
+    "producer_batch_seconds": 0.0725863401718101,
+    "producer_p95_seconds": 0.18913450299987744,
+    "gpu_update_seconds": 0.973784668791603,
+    "producer_batches_per_second": 13.776696794920703,
+    "gpu_batches_per_second": 1.026921076135783,
+    "producer_headroom": 13.415536125484195,
     "sustainable": true,
-    "buffer_seconds": 60.22212908357221,
+    "buffer_seconds": 60.374649465079386,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423892.98836539994,
-    "upload_bytes_per_second": 5757128.6527633555,
+    "generation_bytes_per_second": 429094.8425280772,
+    "upload_bytes_per_second": 5670437.599585741,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.95756757947444,
+    "estimated_full_shards_per_hour": 92.07376438981758,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 12078751,
       "pages": 37,
       "shards": 5
+    },
+    {
+      "seconds": 0.9161946559997887,
+      "bytes": 856768,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -78,3 +78,4 @@
 {"step": 362800, "loss": 0.5554407816380262, "lr": 0.0003, "grad_norm": 0.21308853438439007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53103702390, "tokens_per_second": 151021.7899532143, "data_wait_s": 1.0656998711056076e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 362900, "loss": 0.6023930348455906, "lr": 0.0003, "grad_norm": 0.16299847025574002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53118379197, "tokens_per_second": 150001.53544897021, "data_wait_s": 1.1322999853291549e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 363000, "loss": 0.5643182136118412, "lr": 0.0003, "grad_norm": 0.17398957627821973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53132918813, "tokens_per_second": 148289.47770744606, "data_wait_s": 9.016999683808535e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 363100, "loss": 0.6034362055361271, "lr": 0.0003, "grad_norm": 0.16262701654764686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53147518329, "tokens_per_second": 149494.26927566796, "data_wait_s": 1.5560999599983916e-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 sha256:82e91f29315ce83271b9d760482aa94197497a52f34def71ee98d1607e83b0fb
-size 170388
+oid sha256:197d6f6daa455cd9f68ca30fc4440435a3cf8ae8cafc64e251a8bf157c3e7b9c
+size 172506