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": 75502.79806401806,
-      "required_pages_per_second_upper_estimate": 0.0065443911760968915
+      "required_token_positions_per_second": 75475.66797605852,
+      "required_pages_per_second_upper_estimate": 0.006542039608833138
     },
     {
       "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": 9437.849758002258,
-      "required_pages_per_second_upper_estimate": 0.0037648993768957468
+      "required_token_positions_per_second": 9434.458497007316,
+      "required_pages_per_second_upper_estimate": 0.003763546552180994
     },
     {
       "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": 18875.699516004515,
-      "required_pages_per_second_upper_estimate": 0.004366560882062235
+      "required_token_positions_per_second": 18868.91699401463,
+      "required_pages_per_second_upper_estimate": 0.004364991864968185
     },
     {
       "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": 28313.549274006775,
-      "required_pages_per_second_upper_estimate": 0.0017241466010744985
+      "required_token_positions_per_second": 28303.375491021947,
+      "required_pages_per_second_upper_estimate": 0.0017235270710684366
     },
     {
       "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": 28313.549274006775,
-      "required_pages_per_second_upper_estimate": 0.002179655400033778
+      "required_token_positions_per_second": 28303.375491021947,
+      "required_pages_per_second_upper_estimate": 0.002178872194056772
     },
     {
       "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": 28313.549274006775,
-      "required_pages_per_second_upper_estimate": 0.35155952262943463
+      "required_token_positions_per_second": 28303.375491021947,
+      "required_pages_per_second_upper_estimate": 0.35143319829422437
     }
   ],
   "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": 0.4261239306796085,
-    "producer_p95_seconds": 0.3102632200025255,
-    "gpu_update_seconds": 0.9764936120312623,
-    "producer_batches_per_second": 2.3467351350231347,
-    "gpu_batches_per_second": 1.0240722393665644,
-    "producer_headroom": 2.2915718684794126,
+    "producer_batch_seconds": 0.2685658619376454,
+    "producer_p95_seconds": 0.3259135989974311,
+    "gpu_update_seconds": 0.9768446173061641,
+    "producer_batches_per_second": 3.7234814312780236,
+    "gpu_batches_per_second": 1.0237042639981897,
+    "producer_headroom": 3.637262793783389,
     "sustainable": true,
-    "buffer_seconds": 60.54260394593826,
+    "buffer_seconds": 60.56436627298218,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385695.18532543484,
-    "upload_bytes_per_second": 4071480.653287653,
+    "generation_bytes_per_second": 372713.72508493555,
+    "upload_bytes_per_second": 4089342.977418014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.7612082464436,
+    "estimated_full_shards_per_hour": 79.97568907176066,
     "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": 1.2958799169973645,
-      "bytes": 6217387,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9571121369990578,
       "bytes": 4079115,
@@ -562,6 +556,12 @@
       "bytes": 5751347,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.0857413720004843,
+      "bytes": 7211440,
+      "pages": 43,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -265,3 +265,4 @@
 {"step": 341500, "loss": 0.6143929120153189, "lr": 0.0003, "grad_norm": 0.19491709326595896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49986554279, "tokens_per_second": 151903.6701715982, "data_wait_s": 1.0818999726325274e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341600, "loss": 0.5268503502011299, "lr": 0.0003, "grad_norm": 0.18204988849407544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50001276529, "tokens_per_second": 149005.6634609831, "data_wait_s": 1.3790999219054356e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341700, "loss": 0.6026564370840788, "lr": 0.0003, "grad_norm": 0.14756879543224583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50015922048, "tokens_per_second": 149162.0375101352, "data_wait_s": 1.888599945232272e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 341800, "loss": 0.5851886924356222, "lr": 0.0003, "grad_norm": 0.1859805109652498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50030654863, "tokens_per_second": 149067.29383108535, "data_wait_s": 1.384899951517582e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:209df16919efd86a62737a26a7b4ca4823998152facb318c500ddf2cc321846e
-size 569586
+oid sha256:e461190b064d89f78c0957e3ff858a9b175c92e568b831c42314827e70743416
+size 571704