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": 77498.14284374249,
-      "required_pages_per_second_upper_estimate": 0.005361501342395929
+      "required_token_positions_per_second": 77328.50784539769,
+      "required_pages_per_second_upper_estimate": 0.005349765599603011
     },
     {
       "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": 9687.267855467811,
-      "required_pages_per_second_upper_estimate": 0.0040558567579280285
+      "required_token_positions_per_second": 9666.063480674711,
+      "required_pages_per_second_upper_estimate": 0.004046978928999856
     },
     {
       "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": 19374.535710935623,
-      "required_pages_per_second_upper_estimate": 0.004035204847721908
+      "required_token_positions_per_second": 19332.126961349422,
+      "required_pages_per_second_upper_estimate": 0.004026372223577037
     },
     {
       "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": 29061.803566403436,
-      "required_pages_per_second_upper_estimate": 0.0017422107490012247
+      "required_token_positions_per_second": 28998.190442024134,
+      "required_pages_per_second_upper_estimate": 0.0017383972393263023
     },
     {
       "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": 29061.803566403436,
-      "required_pages_per_second_upper_estimate": 0.002259427397923402
+      "required_token_positions_per_second": 28998.190442024134,
+      "required_pages_per_second_upper_estimate": 0.002254481757307475
     },
     {
       "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": 29061.803566403436,
-      "required_pages_per_second_upper_estimate": 0.3727736120163086
+      "required_token_positions_per_second": 28998.190442024134,
+      "required_pages_per_second_upper_estimate": 0.37195765115922236
     }
   ],
   "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": 0.10962806506222478,
-    "producer_p95_seconds": 0.3013270369992824,
-    "gpu_update_seconds": 0.9513518297935973,
-    "producer_batches_per_second": 9.121751801716112,
-    "gpu_batches_per_second": 1.0511358350116984,
-    "producer_headroom": 8.677995267485667,
+    "producer_batch_seconds": 0.08565658115617225,
+    "producer_p95_seconds": 0.2982594299974153,
+    "gpu_update_seconds": 0.9534388035445329,
+    "producer_batches_per_second": 11.674526189374323,
+    "gpu_batches_per_second": 1.0488350130940443,
+    "producer_headroom": 11.130946281946368,
     "sustainable": true,
-    "buffer_seconds": 60.88651710679023,
+    "buffer_seconds": 60.06664462330557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 545709.3025308239,
-    "upload_bytes_per_second": 5551890.467088929,
+    "generation_bytes_per_second": 552065.8090475652,
+    "upload_bytes_per_second": 5424815.522141014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.09651286071336,
+    "estimated_full_shards_per_hour": 118.46047118730752,
     "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": 1.2739770709995355,
-      "bytes": 15491147,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.0505857130010554,
       "bytes": 2636972,
@@ -562,6 +556,12 @@
       "bytes": 6157279,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.4413105299972813,
+      "bytes": 6425996,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -346,3 +346,4 @@
 {"step": 679600, "loss": 0.5609444975852966, "lr": 0.0001, "grad_norm": 0.18971551553261026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99476059823, "tokens_per_second": 153989.99817463756, "data_wait_s": 9.885996405500919e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 679700, "loss": 0.5336342453956604, "lr": 0.0001, "grad_norm": 0.19925302230334643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99490709871, "tokens_per_second": 152641.51544627736, "data_wait_s": 1.5267003618646413e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679800, "loss": 0.5490105152130127, "lr": 0.0001, "grad_norm": 0.1900582338257386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99505223553, "tokens_per_second": 151373.19111403343, "data_wait_s": 1.436899765394628e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 679900, "loss": 0.5150192379951477, "lr": 0.0001, "grad_norm": 0.19555169141593992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99519707470, "tokens_per_second": 150709.15686407886, "data_wait_s": 9.866002073977143e-06, "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 sha256:9cc7ad879cf2ae400f6a0ef52279b8e4bc65a5c46d0289a2a3d993a45b19dfa2
-size 741927
+oid sha256:ebec0f4b949b318864d89faf24f69a2ab95f306b8fc30da3ad40db97a2546bed
+size 744045