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": 77375.44446388917,
-      "required_pages_per_second_upper_estimate": 0.005353012783778211
+      "required_token_positions_per_second": 77332.37287374001,
+      "required_pages_per_second_upper_estimate": 0.005350032991231833
     },
     {
       "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": 9671.930557986147,
-      "required_pages_per_second_upper_estimate": 0.0040494353517516474
+      "required_token_positions_per_second": 9666.546609217501,
+      "required_pages_per_second_upper_estimate": 0.004047181204831851
     },
     {
       "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": 19343.861115972293,
-      "required_pages_per_second_upper_estimate": 0.004028816138534515
+      "required_token_positions_per_second": 19333.093218435002,
+      "required_pages_per_second_upper_estimate": 0.004026573469446051
     },
     {
       "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": 29015.79167395844,
-      "required_pages_per_second_upper_estimate": 0.0017394524063052389
+      "required_token_positions_per_second": 28999.639827652503,
+      "required_pages_per_second_upper_estimate": 0.0017384841277816394
     },
     {
       "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": 29015.79167395844,
-      "required_pages_per_second_upper_estimate": 0.00225585017567073
+      "required_token_positions_per_second": 28999.639827652503,
+      "required_pages_per_second_upper_estimate": 0.002254594440664907
     },
     {
       "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": 29015.79167395844,
-      "required_pages_per_second_upper_estimate": 0.37218342086374523
+      "required_token_positions_per_second": 28999.639827652503,
+      "required_pages_per_second_upper_estimate": 0.371976242321834
     }
   ],
   "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": 0.08362319339855162,
-    "producer_p95_seconds": 0.29662438800005475,
-    "gpu_update_seconds": 0.9528604392625954,
-    "producer_batches_per_second": 11.958404831946066,
-    "gpu_batches_per_second": 1.0494716317259274,
-    "producer_headroom": 11.39469088104807,
+    "producer_batch_seconds": 0.1019104930938397,
+    "producer_p95_seconds": 0.294245918001252,
+    "gpu_update_seconds": 0.9533911512113453,
+    "producer_batches_per_second": 9.812532249051085,
+    "gpu_batches_per_second": 1.0488874358959963,
+    "producer_headroom": 9.355181417221266,
     "sustainable": true,
-    "buffer_seconds": 60.030207673543515,
+    "buffer_seconds": 60.06364252631476,
     "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": 308465.0115167414,
-    "upload_bytes_per_second": 5509686.916635343,
+    "generation_bytes_per_second": 306968.7059050282,
+    "upload_bytes_per_second": 5543887.316173678,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.18941077353175,
+    "estimated_full_shards_per_hour": 65.86833842147001,
     "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": 1.482665261002694,
-      "bytes": 2527538,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.213385584000207,
       "bytes": 6122336,
@@ -562,6 +556,12 @@
       "bytes": 2650389,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.6456713849984226,
+      "bytes": 6071875,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -317,3 +317,4 @@
 {"step": 676700, "loss": 0.5441694855690002, "lr": 0.0001, "grad_norm": 0.18920825808135824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99051399436, "tokens_per_second": 151587.82029818394, "data_wait_s": 1.758299913490191e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 676800, "loss": 0.5175274014472961, "lr": 0.0001, "grad_norm": 0.19439754819100055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99066050054, "tokens_per_second": 152579.46254578032, "data_wait_s": 2.260000110254623e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 676900, "loss": 0.5266819596290588, "lr": 0.0001, "grad_norm": 0.2042373696576169, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99080750240, "tokens_per_second": 152655.05764364393, "data_wait_s": 1.166099900729023e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 677000, "loss": 0.6073641180992126, "lr": 0.0001, "grad_norm": 0.20068715210851132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99095332637, "tokens_per_second": 146804.93471253722, "data_wait_s": 1.1740998161258176e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de24077bafce4a043b6c94e79f72b29338ae3285dbc6a55a1a405862ed77745b
-size 680505
+oid sha256:bd9da277dab3d6db77c92be83a5a69291418657a7d52aa83870e91a512211bcf
+size 682623