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": 77520.38094113626,
-      "required_pages_per_second_upper_estimate": 0.005363039825573125
+      "required_token_positions_per_second": 76608.11503721995,
+      "required_pages_per_second_upper_estimate": 0.005299927153591666
     },
     {
       "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": 9690.047617642032,
-      "required_pages_per_second_upper_estimate": 0.00405702058630234
+      "required_token_positions_per_second": 9576.014379652494,
+      "required_pages_per_second_upper_estimate": 0.004009277250840914
     },
     {
       "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": 19380.095235284065,
-      "required_pages_per_second_upper_estimate": 0.0040363627500291715
+      "required_token_positions_per_second": 19152.028759304987,
+      "required_pages_per_second_upper_estimate": 0.00398886251760016
     },
     {
       "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": 29070.1428529261,
-      "required_pages_per_second_upper_estimate": 0.001742710676494905
+      "required_token_positions_per_second": 28728.04313895748,
+      "required_pages_per_second_upper_estimate": 0.0017222023209985042
     },
     {
       "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": 29070.1428529261,
-      "required_pages_per_second_upper_estimate": 0.0022600757407698945
+      "required_token_positions_per_second": 28728.04313895748,
+      "required_pages_per_second_upper_estimate": 0.0022334789927464507
     },
     {
       "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": 29070.1428529261,
-      "required_pages_per_second_upper_estimate": 0.3728805794297931
+      "required_token_positions_per_second": 28728.04313895748,
+      "required_pages_per_second_upper_estimate": 0.36849249161705827
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08835190990635056,
-    "producer_p95_seconds": 0.3068851400021231,
-    "gpu_update_seconds": 0.9510789176330811,
-    "producer_batches_per_second": 11.318374453477684,
-    "gpu_batches_per_second": 1.0514374585115052,
-    "producer_headroom": 10.76466732457947,
+    "producer_batch_seconds": 0.11191009941398988,
+    "producer_p95_seconds": 0.3243526459991699,
+    "gpu_update_seconds": 0.9624045698576366,
+    "producer_batches_per_second": 8.935744005558359,
+    "gpu_batches_per_second": 1.0390640602921541,
+    "producer_headroom": 8.599800866027346,
     "sustainable": true,
-    "buffer_seconds": 60.86905072851719,
+    "buffer_seconds": 60.6314879010311,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382436.5258028965,
-    "upload_bytes_per_second": 4798865.727477853,
+    "generation_bytes_per_second": 338247.8083685125,
+    "upload_bytes_per_second": 4786828.827312803,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.06197577061816,
+    "estimated_full_shards_per_hour": 72.58010566989451,
     "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": 1.2924744820011256,
-      "bytes": 2413023,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0274238659985713,
       "bytes": 6279174,
@@ -562,6 +556,12 @@
       "bytes": 2429729,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1435100689996034,
+      "bytes": 625562,
+      "pages": 19,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -355,3 +355,4 @@
 {"step": 795500, "loss": 0.4830298125743866, "lr": 3e-05, "grad_norm": 0.26718392901315674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116438443906, "tokens_per_second": 152073.72058413352, "data_wait_s": 8.909999451134354e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 795600, "loss": 0.5383407473564148, "lr": 3e-05, "grad_norm": 0.2748085125557036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116453264956, "tokens_per_second": 154613.32209299205, "data_wait_s": 9.90999978967011e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 795700, "loss": 0.6049256920814514, "lr": 3e-05, "grad_norm": 0.28892878948135853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116468097835, "tokens_per_second": 126994.52916165971, "data_wait_s": 9.45699866861105e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 795800, "loss": 0.5575957894325256, "lr": 3e-05, "grad_norm": 0.27267736367051293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116482759554, "tokens_per_second": 151456.69218906824, "data_wait_s": 8.636001439299434e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64ae6f04c60175c9cedc6a476830a53c2d6f7fb273e7383e6960992d5c02dc2e
-size 761772
+oid sha256:e52bbe18786a68e199b436a88dda072c21e9b893ae66c9cee6d09d2359850a7a
+size 763890