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": 75654.98803536229,
-      "required_pages_per_second_upper_estimate": 0.006557582617091053
+      "required_token_positions_per_second": 75578.13493590128,
+      "required_pages_per_second_upper_estimate": 0.006550921185212172
     },
     {
       "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": 9456.873504420286,
-      "required_pages_per_second_upper_estimate": 0.0037724882337722538
+      "required_token_positions_per_second": 9447.26686698766,
+      "required_pages_per_second_upper_estimate": 0.00376865600246835
     },
     {
       "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": 18913.74700884057,
-      "required_pages_per_second_upper_estimate": 0.0043753624999168525
+      "required_token_positions_per_second": 18894.53373397532,
+      "required_pages_per_second_upper_estimate": 0.004370917846918865
     },
     {
       "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": 28370.620513260856,
-      "required_pages_per_second_upper_estimate": 0.0017276219401154208
+      "required_token_positions_per_second": 28341.80060096298,
+      "required_pages_per_second_upper_estimate": 0.001725866958662874
     },
     {
       "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": 28370.620513260856,
-      "required_pages_per_second_upper_estimate": 0.0021840489020148587
+      "required_token_positions_per_second": 28341.80060096298,
+      "required_pages_per_second_upper_estimate": 0.002181830265387545
     },
     {
       "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": 28370.620513260856,
-      "required_pages_per_second_upper_estimate": 0.3522681564158195
+      "required_token_positions_per_second": 28341.80060096298,
+      "required_pages_per_second_upper_estimate": 0.3519103095591216
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.09918981006251215,
-    "producer_p95_seconds": 0.2953344889974687,
-    "gpu_update_seconds": 0.9745292665374347,
-    "producer_batches_per_second": 10.081680763072058,
-    "gpu_batches_per_second": 1.0261364479622705,
-    "producer_headroom": 9.824892959501177,
+    "producer_batch_seconds": 0.2573587191795923,
+    "producer_p95_seconds": 0.33661060400118004,
+    "gpu_update_seconds": 0.9755202356148323,
+    "producer_batches_per_second": 3.8856270469009107,
+    "gpu_batches_per_second": 1.0250940610880708,
+    "producer_headroom": 3.790507812304141,
     "sustainable": true,
-    "buffer_seconds": 60.42081452532095,
+    "buffer_seconds": 60.4822546081196,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 162,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 630359.4242213133,
-    "upload_bytes_per_second": 5005455.9790586885,
+    "generation_bytes_per_second": 583520.7425960273,
+    "upload_bytes_per_second": 4965514.8487608,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.63022921075606,
+    "estimated_full_shards_per_hour": 62.604983846715044,
     "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": 1.6352041230020404,
-      "bytes": 7545726,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.4504600340005709,
       "bytes": 11937829,
@@ -562,6 +556,12 @@
       "bytes": 11880694,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.2683707119977043,
+      "bytes": 2325676,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 382800, "loss": 0.5965701099485159, "lr": 0.0003, "grad_norm": 0.15886674502314926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56030838053, "tokens_per_second": 151526.8503122318, "data_wait_s": 1.5002999134594575e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382900, "loss": 0.6087070219218731, "lr": 0.0003, "grad_norm": 0.1726813551557998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56045689617, "tokens_per_second": 151337.44953420456, "data_wait_s": 9.678999049356207e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 383000, "loss": 0.543660243973136, "lr": 0.0003, "grad_norm": 0.19995132168341503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56060283476, "tokens_per_second": 148668.3106738177, "data_wait_s": 1.178499951492995e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 383100, "loss": 0.5857820343226194, "lr": 0.0003, "grad_norm": 0.15258547212936122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56074807961, "tokens_per_second": 148263.606208162, "data_wait_s": 1.1571999493753538e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af1e673c0255a052c4000b714a0d80ff9a6afa2b1062997c8afce8b2f9b0ab65
-size 597120
+oid sha256:0d1b2eaf13c523595e8d27dab14a9146778897231e9aa3893918681c32557ff6
+size 599238