Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76578.2994751791,
-      "required_pages_per_second_upper_estimate": 0.005297864443828044
+      "required_token_positions_per_second": 77053.4169452399,
+      "required_pages_per_second_upper_estimate": 0.005330734172831265
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9572.287434397387,
-      "required_pages_per_second_upper_estimate": 0.004007716856690068
+      "required_token_positions_per_second": 9631.677118154987,
+      "required_pages_per_second_upper_estimate": 0.0040325820770817505
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19144.574868794774,
-      "required_pages_per_second_upper_estimate": 0.003987310068779195
+      "required_token_positions_per_second": 19263.354236309973,
+      "required_pages_per_second_upper_estimate": 0.004012048678610043
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28716.862303192163,
-      "required_pages_per_second_upper_estimate": 0.0017215320469665212
+      "required_token_positions_per_second": 28895.03135446496,
+      "required_pages_per_second_upper_estimate": 0.0017322130095419363
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28716.862303192163,
-      "required_pages_per_second_upper_estimate": 0.0022326097319449963
+      "required_token_positions_per_second": 28895.03135446496,
+      "required_pages_per_second_upper_estimate": 0.002246461591998622
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28716.862303192163,
-      "required_pages_per_second_upper_estimate": 0.3683490758609069
+      "required_token_positions_per_second": 28895.03135446496,
+      "required_pages_per_second_upper_estimate": 0.3706344371476118
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1316351952812056,
-    "producer_p95_seconds": 0.36703890100034187,
-    "gpu_update_seconds": 0.9627792795777222,
-    "producer_batches_per_second": 7.596752508808534,
-    "gpu_batches_per_second": 1.038659660850411,
-    "producer_headroom": 7.313995907560935,
+    "producer_batch_seconds": 0.5451187155549064,
+    "producer_p95_seconds": 0.378449236995948,
+    "gpu_update_seconds": 0.956842706305897,
+    "producer_batches_per_second": 1.8344627903337438,
+    "gpu_batches_per_second": 1.0451038539664699,
+    "producer_headroom": 1.7552923409204066,
     "sustainable": true,
-    "buffer_seconds": 60.6550946133965,
+    "buffer_seconds": 60.28109049727151,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352655.9844869595,
-    "upload_bytes_per_second": 4691144.814148951,
+    "generation_bytes_per_second": 313096.0479355008,
+    "upload_bytes_per_second": 4689555.703688069,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.6717648597392,
+    "estimated_full_shards_per_hour": 67.18312338398712,
     "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": 1.1005420340006822,
-      "bytes": 1290047,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.504874573998677,
       "bytes": 12022604,
@@ -562,6 +556,12 @@
       "bytes": 6408361,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3097595139988698,
+      "bytes": 2126792,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -371,3 +371,4 @@
 {"step": 641500, "loss": 0.5121535062789917, "lr": 0.0001, "grad_norm": 0.1780030994972999, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93899177171, "tokens_per_second": 152155.189234055, "data_wait_s": 1.7579004634171724e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 641600, "loss": 0.5751965641975403, "lr": 0.0001, "grad_norm": 0.17172272387595733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93913834487, "tokens_per_second": 151922.87912933656, "data_wait_s": 1.0546005796641111e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 641700, "loss": 0.5535345673561096, "lr": 0.0001, "grad_norm": 0.16103962629784696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93928381119, "tokens_per_second": 151702.892769641, "data_wait_s": 1.2272997992113233e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 641800, "loss": 0.5942527055740356, "lr": 0.0001, "grad_norm": 0.1982612733062987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93942907709, "tokens_per_second": 129289.82179239734, "data_wait_s": 1.2977994629181921e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d21c78744643607df2cda9045c4a895f54b49a5114914607e04b9494b019df99
-size 796443
+oid sha256:1a87d0a6c565671b1ef8c871fa51b02512ec72ddf8d1c8ac3099903f2ed83fc6
+size 798561