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": 75620.74897542187,
-      "required_pages_per_second_upper_estimate": 0.006554614862153522
+      "required_token_positions_per_second": 75280.5920841482,
+      "required_pages_per_second_upper_estimate": 0.006525130924937679
     },
     {
       "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": 9452.593621927734,
-      "required_pages_per_second_upper_estimate": 0.003770780924656029
+      "required_token_positions_per_second": 9410.074010518525,
+      "required_pages_per_second_upper_estimate": 0.003753819215940053
     },
     {
       "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": 18905.187243855467,
-      "required_pages_per_second_upper_estimate": 0.004373382348934267
+      "required_token_positions_per_second": 18820.14802103705,
+      "required_pages_per_second_upper_estimate": 0.004353710021374431
     },
     {
       "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": 28357.7808657832,
-      "required_pages_per_second_upper_estimate": 0.0017268400729484558
+      "required_token_positions_per_second": 28230.22203155558,
+      "required_pages_per_second_upper_estimate": 0.0017190724091934746
     },
     {
       "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": 28357.7808657832,
-      "required_pages_per_second_upper_estimate": 0.00218306047040961
+      "required_token_positions_per_second": 28230.22203155558,
+      "required_pages_per_second_upper_estimate": 0.0021732406382453146
     },
     {
       "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": 28357.7808657832,
-      "required_pages_per_second_upper_estimate": 0.35210873096568285
+      "required_token_positions_per_second": 28230.22203155558,
+      "required_pages_per_second_upper_estimate": 0.3505248771565315
     }
   ],
   "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": 0.29809513047649716,
-    "producer_p95_seconds": 0.3465591649983253,
-    "gpu_update_seconds": 0.9749705074193719,
-    "producer_batches_per_second": 3.354633798953799,
-    "gpu_batches_per_second": 1.0256720509904225,
-    "producer_headroom": 3.2706690171721604,
+    "producer_batch_seconds": 0.10530740821084805,
+    "producer_p95_seconds": 0.3032622579994495,
+    "gpu_update_seconds": 0.9793759315493596,
+    "producer_batches_per_second": 9.496008087083343,
+    "gpu_batches_per_second": 1.0210583778774442,
+    "producer_headroom": 9.300161766287502,
     "sustainable": true,
-    "buffer_seconds": 60.448171460001056,
+    "buffer_seconds": 60.7213077560603,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290834.8809515413,
-    "upload_bytes_per_second": 5050172.753931662,
+    "generation_bytes_per_second": 261016.62828296778,
+    "upload_bytes_per_second": 5153059.332158155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.406395162674706,
+    "estimated_full_shards_per_hour": 56.0080922733953,
     "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": 1.1101788160012802,
-      "bytes": 5690031,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.5474547390003863,
       "bytes": 13326359,
@@ -562,6 +556,12 @@
       "bytes": 5852988,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.7732943640003214,
+      "bytes": 17450620,
+      "pages": 33,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 261000, "loss": 0.5985351726412773, "lr": 0.0003, "grad_norm": 0.14187585540698935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38203433879, "tokens_per_second": 149347.4343050983, "data_wait_s": 1.337999856332317e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261100, "loss": 0.5764923933893442, "lr": 0.0003, "grad_norm": 0.14409881021038268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38218086483, "tokens_per_second": 149397.42575111936, "data_wait_s": 1.006700040306896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261200, "loss": 0.6021216567605734, "lr": 0.0003, "grad_norm": 0.16239392872977046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38232537918, "tokens_per_second": 127206.86807270769, "data_wait_s": 1.2266998965060338e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 261300, "loss": 0.5932094510644674, "lr": 0.0003, "grad_norm": 0.16197659898215175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38247263359, "tokens_per_second": 150637.85039217872, "data_wait_s": 1.0947998816845939e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10e13e60c581755e448c088e923690475ffbb16c5e3c7bd153163077ba2a267f
-size 514518
+oid sha256:a9fb24769f86eac2c6614066c2c685abc74914a64ac6501f256d9f6840c1bb1c
+size 516636