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": 77378.44019359797,
-      "required_pages_per_second_upper_estimate": 0.0053532200353105155
+      "required_token_positions_per_second": 76712.94137364216,
+      "required_pages_per_second_upper_estimate": 0.005307179282775965
     },
     {
       "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": 9672.305024199746,
-      "required_pages_per_second_upper_estimate": 0.004049592132935538
+      "required_token_positions_per_second": 9589.11767170527,
+      "required_pages_per_second_upper_estimate": 0.0040147633255955585
     },
     {
       "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": 19344.610048399492,
-      "required_pages_per_second_upper_estimate": 0.004028972121408424
+      "required_token_positions_per_second": 19178.23534341054,
+      "required_pages_per_second_upper_estimate": 0.003994320657955248
     },
     {
       "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": 29016.91507259924,
-      "required_pages_per_second_upper_estimate": 0.0017395197523384245
+      "required_token_positions_per_second": 28767.35301511581,
+      "required_pages_per_second_upper_estimate": 0.0017245588880514906
     },
     {
       "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": 29016.91507259924,
-      "required_pages_per_second_upper_estimate": 0.0022559375149737437
+      "required_token_positions_per_second": 28767.35301511581,
+      "required_pages_per_second_upper_estimate": 0.0022365351627118887
     },
     {
       "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": 29016.91507259924,
-      "required_pages_per_second_upper_estimate": 0.37219783061529793
+      "required_token_positions_per_second": 28767.35301511581,
+      "required_pages_per_second_upper_estimate": 0.36899671650076077
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3683819505233856,
-    "producer_p95_seconds": 0.18601086900071095,
-    "gpu_update_seconds": 0.9528235489825757,
-    "producer_batches_per_second": 2.7145738236610972,
-    "gpu_batches_per_second": 1.0495122639105627,
-    "producer_headroom": 2.586509864635967,
+    "producer_batch_seconds": 0.05458811764064109,
+    "producer_p95_seconds": 0.17259476100025495,
+    "gpu_update_seconds": 0.9610894678239029,
+    "producer_batches_per_second": 18.3190050000093,
+    "gpu_batches_per_second": 1.040485858474964,
+    "producer_headroom": 17.606202766522355,
     "sustainable": true,
-    "buffer_seconds": 60.027883585902266,
+    "buffer_seconds": 60.54863647290588,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372337.0480323842,
-    "upload_bytes_per_second": 4489845.021782463,
+    "generation_bytes_per_second": 357380.9093811584,
+    "upload_bytes_per_second": 4520993.195791211,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.89486294487614,
+    "estimated_full_shards_per_hour": 76.68562375141205,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 7642707,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.8502533870005209,
+      "bytes": 10875382,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -68,3 +68,4 @@
 {"step": 849300, "loss": 0.5257542133331299, "lr": 3e-05, "grad_norm": 0.29004753416830137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124313939588, "tokens_per_second": 151700.67123026052, "data_wait_s": 1.1265000466664787e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 849400, "loss": 0.5637456178665161, "lr": 3e-05, "grad_norm": 0.3377127396603084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124328640350, "tokens_per_second": 141248.01703006861, "data_wait_s": 9.518999831925612e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 849500, "loss": 0.5055427551269531, "lr": 3e-05, "grad_norm": 0.29385733964605903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124343326771, "tokens_per_second": 152961.74688759536, "data_wait_s": 1.499599966336973e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 849600, "loss": 0.48633232712745667, "lr": 3e-05, "grad_norm": 0.2856631745968521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124358058707, "tokens_per_second": 153195.3083980939, "data_wait_s": 1.4041000213182997e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee72ae9a16f3bc654f7186ddb4427c16252996a2cbb928324e37b7ce40609d9a
-size 149208
+oid sha256:dc66bc37d1eeed86a7b0c452f3b1b71c588f71cba977d5117da0544aaa10c0bd
+size 151326