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": 75381.97392836728,
-      "required_pages_per_second_upper_estimate": 0.006533918446244654
+      "required_token_positions_per_second": 75277.03724224916,
+      "required_pages_per_second_upper_estimate": 0.006524822800251535
     },
     {
       "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": 9422.74674104591,
-      "required_pages_per_second_upper_estimate": 0.0037588745576216335
+      "required_token_positions_per_second": 9409.629655281145,
+      "required_pages_per_second_upper_estimate": 0.0037536419559921594
     },
     {
       "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": 18845.49348209182,
-      "required_pages_per_second_upper_estimate": 0.004359573247724573
+      "required_token_positions_per_second": 18819.25931056229,
+      "required_pages_per_second_upper_estimate": 0.004353504433846866
     },
     {
       "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": 28268.24022313773,
-      "required_pages_per_second_upper_estimate": 0.001721387517594792
+      "required_token_positions_per_second": 28228.88896584344,
+      "required_pages_per_second_upper_estimate": 0.0017189912324856603
     },
     {
       "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": 28268.24022313773,
-      "required_pages_per_second_upper_estimate": 0.002176167383874399
+      "required_token_positions_per_second": 28228.88896584344,
+      "required_pages_per_second_upper_estimate": 0.0021731380151566323
     },
     {
       "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": 28268.24022313773,
-      "required_pages_per_second_upper_estimate": 0.35099693585727965
+      "required_token_positions_per_second": 28228.88896584344,
+      "required_pages_per_second_upper_estimate": 0.3505083249418707
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.055744199320269416,
-    "producer_p95_seconds": 0.17564570499962429,
-    "gpu_update_seconds": 0.9780587607066513,
-    "producer_batches_per_second": 17.939086258189114,
-    "gpu_batches_per_second": 1.0224334571447384,
-    "producer_headroom": 17.545480473894163,
+    "producer_batch_seconds": 0.10604240690624067,
+    "producer_p95_seconds": 0.17803285800073354,
+    "gpu_update_seconds": 0.9794221810661303,
+    "producer_batches_per_second": 9.43018957391422,
+    "gpu_batches_per_second": 1.0210101622483883,
+    "producer_headroom": 9.236136840350147,
     "sustainable": true,
-    "buffer_seconds": 60.63964316381238,
+    "buffer_seconds": 60.72417522610008,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417396.4747164421,
-    "upload_bytes_per_second": 5134078.494695147,
+    "generation_bytes_per_second": 452530.48358807917,
+    "upload_bytes_per_second": 5235226.743210096,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.56356698150584,
+    "estimated_full_shards_per_hour": 97.1025074074915,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 6142145,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2391148739998243,
+      "bytes": 7141661,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 158500, "loss": 0.6143712177872658, "lr": 0.0003, "grad_norm": 0.15635928735291174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23197315445, "tokens_per_second": 149684.81177275773, "data_wait_s": 1.6041999515437055e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158600, "loss": 0.636771772056818, "lr": 0.0003, "grad_norm": 0.14898210905915524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23211903884, "tokens_per_second": 148077.36626096538, "data_wait_s": 1.3330999536265153e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158700, "loss": 0.5854216739535332, "lr": 0.0003, "grad_norm": 0.13415142926651943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23226643792, "tokens_per_second": 149593.5826521523, "data_wait_s": 1.518199951533461e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 158800, "loss": 0.6244642548263073, "lr": 0.0003, "grad_norm": 0.14388045756002935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23241394389, "tokens_per_second": 149268.78273229898, "data_wait_s": 1.7186999684781767e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:694da9f215855c835f87724e3ab49a40c07efaddf7b0b5921c870415167fd8f7
-size 78531
+oid sha256:0e47484537feeb732d67d2ea78532aa340952eb043fc08483d96fbf930f6a90c
+size 80649