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": 76589.3574227098,
-      "required_pages_per_second_upper_estimate": 0.005298629458296192
+      "required_token_positions_per_second": 76784.45374021186,
+      "required_pages_per_second_upper_estimate": 0.005312126674226808
     },
     {
       "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": 9573.669677838725,
-      "required_pages_per_second_upper_estimate": 0.004008295573154431
+      "required_token_positions_per_second": 9598.056717526482,
+      "required_pages_per_second_upper_estimate": 0.004018505917412396
     },
     {
       "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": 19147.33935567745,
-      "required_pages_per_second_upper_estimate": 0.003987885838492434
+      "required_token_positions_per_second": 19196.113435052965,
+      "required_pages_per_second_upper_estimate": 0.0039980441929673835
     },
     {
       "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": 28721.009033516173,
-      "required_pages_per_second_upper_estimate": 0.0017217806371177548
+      "required_token_positions_per_second": 28794.170152579445,
+      "required_pages_per_second_upper_estimate": 0.0017261665344950375
     },
     {
       "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": 28721.009033516173,
-      "required_pages_per_second_upper_estimate": 0.0022329321219881487
+      "required_token_positions_per_second": 28794.170152579445,
+      "required_pages_per_second_upper_estimate": 0.0022386200771935647
     },
     {
       "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": 28721.009033516173,
-      "required_pages_per_second_upper_estimate": 0.36840226566509116
+      "required_token_positions_per_second": 28794.170152579445,
+      "required_pages_per_second_upper_estimate": 0.36934069794614544
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1193529816794694,
-    "producer_p95_seconds": 0.27461079099884955,
-    "gpu_update_seconds": 0.9626402738057003,
-    "producer_batches_per_second": 8.378508738772597,
-    "gpu_batches_per_second": 1.0388096438627088,
-    "producer_headroom": 8.065489946375505,
+    "producer_batch_seconds": 0.1031627557111392,
+    "producer_p95_seconds": 0.269750429000851,
+    "gpu_update_seconds": 0.9601943675922617,
+    "producer_batches_per_second": 9.693420780655078,
+    "gpu_batches_per_second": 1.041455807023273,
+    "producer_headroom": 9.30756803628679,
     "sustainable": true,
-    "buffer_seconds": 60.646337249759114,
+    "buffer_seconds": 60.49224515831249,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 520769.6505507268,
-    "upload_bytes_per_second": 5116600.151679697,
+    "generation_bytes_per_second": 521367.95228954824,
+    "upload_bytes_per_second": 5178801.942910159,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.74504411116936,
+    "estimated_full_shards_per_hour": 111.87342573656879,
     "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": 1.1081106500005262,
-      "bytes": 1056276,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.211192590999417,
       "bytes": 7370118,
@@ -562,6 +556,12 @@
       "bytes": 5986982,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.4115151900005003,
+      "bytes": 7731251,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 742100, "loss": 0.5540927648544312, "lr": 5e-05, "grad_norm": 0.2346659712796483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108622933641, "tokens_per_second": 149812.94862111897, "data_wait_s": 1.634900036151521e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742200, "loss": 0.5303184986114502, "lr": 5e-05, "grad_norm": 0.23299460527597712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108637634533, "tokens_per_second": 151946.86827446392, "data_wait_s": 1.3856999430572614e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742300, "loss": 0.5082201361656189, "lr": 5e-05, "grad_norm": 0.23735640135720543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108652273678, "tokens_per_second": 151401.48973396877, "data_wait_s": 1.1823998647741973e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 742400, "loss": 0.48355570435523987, "lr": 5e-05, "grad_norm": 0.23324662270761445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108666971916, "tokens_per_second": 150519.64363600602, "data_wait_s": 1.3543998647946864e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5636efa9ac24feabfd19743a5a7ad907e663c48eb3e5ee02b2f24cef5ee76a5
-size 475611
+oid sha256:e19dd5aad81213ce16e457d2ee39b167faa471fec0cd8ca54b500b692448a067
+size 477729