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": 76066.20672241565,
-      "required_pages_per_second_upper_estimate": 0.006593226010660604
+      "required_token_positions_per_second": 75408.59061692265,
+      "required_pages_per_second_upper_estimate": 0.0065362255133492525
     },
     {
       "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": 9508.275840301956,
-      "required_pages_per_second_upper_estimate": 0.0037929933940888606
+      "required_token_positions_per_second": 9426.073827115331,
+      "required_pages_per_second_upper_estimate": 0.003760201781999095
     },
     {
       "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": 19016.551680603912,
-      "required_pages_per_second_upper_estimate": 0.004399144551428874
+      "required_token_positions_per_second": 18852.147654230663,
+      "required_pages_per_second_upper_estimate": 0.004361112573339464
     },
     {
       "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": 28524.827520905867,
-      "required_pages_per_second_upper_estimate": 0.0017370123378193604
+      "required_token_positions_per_second": 28278.22148134599,
+      "required_pages_per_second_upper_estimate": 0.0017219953238520602
     },
     {
       "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": 28524.827520905867,
-      "required_pages_per_second_upper_estimate": 0.00219592018433569
+      "required_token_positions_per_second": 28278.22148134599,
+      "required_pages_per_second_upper_estimate": 0.0021769357687611616
     },
     {
       "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": 28524.827520905867,
-      "required_pages_per_second_upper_estimate": 0.35418289135311554
+      "required_token_positions_per_second": 28278.22148134599,
+      "required_pages_per_second_upper_estimate": 0.3511208696791039
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07829836524996381,
-    "producer_p95_seconds": 0.19687288199929753,
-    "gpu_update_seconds": 0.9692609001662416,
-    "producer_batches_per_second": 12.771658728857844,
-    "gpu_batches_per_second": 1.0317139583660975,
-    "producer_headroom": 12.379069436148791,
+    "producer_batch_seconds": 0.07964664403903043,
+    "producer_p95_seconds": 0.20439933899979223,
+    "gpu_update_seconds": 0.9777135389592403,
+    "producer_batches_per_second": 12.555456818870548,
+    "gpu_batches_per_second": 1.022794469088035,
+    "producer_headroom": 12.27564011962785,
     "sustainable": true,
-    "buffer_seconds": 60.09417581030698,
+    "buffer_seconds": 60.6182394154729,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299521.5781140328,
-    "upload_bytes_per_second": 5569064.774391282,
+    "generation_bytes_per_second": 310794.40576267836,
+    "upload_bytes_per_second": 5486306.114488939,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.27035815778483,
+    "estimated_full_shards_per_hour": 66.68924455318702,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 9926404,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9813094319997617,
+      "bytes": 1049873,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -53,3 +53,4 @@
 {"step": 285500, "loss": 0.6272663809359074, "lr": 0.0003, "grad_norm": 0.14719175495472536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41789375829, "tokens_per_second": 151690.0631004838, "data_wait_s": 1.4142999134492129e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285600, "loss": 0.6442608311772346, "lr": 0.0003, "grad_norm": 0.16393942543920254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41803927346, "tokens_per_second": 131592.77003216252, "data_wait_s": 6.4779997046571225e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 285700, "loss": 0.608401644974947, "lr": 0.0003, "grad_norm": 0.1501933163471568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41818387961, "tokens_per_second": 148086.78253792008, "data_wait_s": 1.215599877468776e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 285800, "loss": 0.553261948749423, "lr": 0.0003, "grad_norm": 0.14661093419425655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41832941344, "tokens_per_second": 148412.1016033091, "data_wait_s": 9.921999662765302e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1f4474e7f4123175d4a2a297632200519d41586607137df66028f3b722f4a27
-size 117438
+oid sha256:8338a06956dd630f1b8f9c450f6d23daaaf91e69e04e6175cad7ff3d6a92974e
+size 119556