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": 75091.48216303253,
-      "required_pages_per_second_upper_estimate": 0.005195002972282279
+      "required_token_positions_per_second": 76738.88823189473,
+      "required_pages_per_second_upper_estimate": 0.005308974346634884
     },
     {
       "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": 9386.435270379066,
-      "required_pages_per_second_upper_estimate": 0.003929904436650109
+      "required_token_positions_per_second": 9592.361028986841,
+      "required_pages_per_second_upper_estimate": 0.004016121251560351
     },
     {
       "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": 18772.870540758133,
-      "required_pages_per_second_upper_estimate": 0.003909893859751482
+      "required_token_positions_per_second": 19184.722057973682,
+      "required_pages_per_second_upper_estimate": 0.003995671669532585
     },
     {
       "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": 28159.305811137197,
-      "required_pages_per_second_upper_estimate": 0.0016881073866072951
+      "required_token_positions_per_second": 28777.083086960527,
+      "required_pages_per_second_upper_estimate": 0.001725142191001622
     },
     {
       "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": 28159.305811137197,
-      "required_pages_per_second_upper_estimate": 0.00218926216711955
+      "required_token_positions_per_second": 28777.083086960527,
+      "required_pages_per_second_upper_estimate": 0.002237291632999752
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28159.305811137197,
-      "required_pages_per_second_upper_estimate": 0.3611973398383448
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28777.083086960527,
+      "required_pages_per_second_upper_estimate": 0.3691215234150476
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.6294189336250327,
-    "producer_p95_seconds": 0.24311506599769928,
-    "gpu_update_seconds": 0.9818423857972034,
-    "producer_batches_per_second": 1.588766950877483,
-    "gpu_batches_per_second": 1.0184934104143952,
-    "producer_headroom": 1.5599187335252962,
+    "producer_batch_seconds": 0.06235414972655917,
+    "producer_p95_seconds": 0.22041695000007167,
+    "gpu_update_seconds": 0.9607645054382827,
+    "producer_batches_per_second": 16.03742500515662,
+    "gpu_batches_per_second": 1.0408377852633293,
+    "producer_headroom": 15.408188703582852,
     "sustainable": true,
-    "buffer_seconds": 60.87422791942661,
+    "buffer_seconds": 60.52816384261181,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344925.38032408245,
-    "upload_bytes_per_second": 1871987.2329020484,
+    "generation_bytes_per_second": 345108.13944355503,
+    "upload_bytes_per_second": 1896617.331969005,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.0129571656404,
+    "estimated_full_shards_per_hour": 74.05217301826465,
     "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": 1.0547332340011053,
-      "bytes": 6465464,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.862101573999098,
       "bytes": 2178443,
@@ -562,6 +556,12 @@
       "bytes": 5373163,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.302073504000873,
+      "bytes": 12612142,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -184,3 +184,4 @@
 {"step": 663400, "loss": 0.5194874405860901, "lr": 0.0001, "grad_norm": 0.17808860347510522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97104298856, "tokens_per_second": 151876.80848047035, "data_wait_s": 9.908002539305016e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 663500, "loss": 0.515264630317688, "lr": 0.0001, "grad_norm": 0.1931068198266987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97119043307, "tokens_per_second": 153308.1898406546, "data_wait_s": 1.1003001418430358e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 663600, "loss": 0.5422067642211914, "lr": 0.0001, "grad_norm": 0.251312046759925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97133724836, "tokens_per_second": 152555.63012917296, "data_wait_s": 1.308799983235076e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 663700, "loss": 0.5779117345809937, "lr": 0.0001, "grad_norm": 0.2095624644082802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97148526991, "tokens_per_second": 154575.90331443175, "data_wait_s": 1.5607998648192734e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c757d5556c246925bf62aa28610f71b5c45d0b66f15ab447df6bfbd4cdf03b22
-size 396462
+oid sha256:a8864e82e311645a30fad52e8fac842a118573d26f15866c0c39db0f79d5bd69
+size 398580