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": 77234.29851115317,
-      "required_pages_per_second_upper_estimate": 0.0053432479792642025
+      "required_token_positions_per_second": 77461.3069774061,
+      "required_pages_per_second_upper_estimate": 0.0053589529516917915
     },
     {
       "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": 9654.287313894147,
-      "required_pages_per_second_upper_estimate": 0.0040420484938831595
+      "required_token_positions_per_second": 9682.663372175763,
+      "required_pages_per_second_upper_estimate": 0.004053928956926193
     },
     {
       "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": 19308.574627788294,
-      "required_pages_per_second_upper_estimate": 0.004021466893613059
+      "required_token_positions_per_second": 19365.326744351525,
+      "required_pages_per_second_upper_estimate": 0.004033286862839221
     },
     {
       "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": 28962.86194168244,
-      "required_pages_per_second_upper_estimate": 0.0017362793496743143
+      "required_token_positions_per_second": 29047.99011652729,
+      "required_pages_per_second_upper_estimate": 0.0017413826537730664
     },
     {
       "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": 28962.86194168244,
-      "required_pages_per_second_upper_estimate": 0.0022517351217995587
+      "required_token_positions_per_second": 29047.99011652729,
+      "required_pages_per_second_upper_estimate": 0.0022583534629544767
     },
     {
       "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": 28962.86194168244,
-      "required_pages_per_second_upper_estimate": 0.3715044950896274
+      "required_token_positions_per_second": 29047.99011652729,
+      "required_pages_per_second_upper_estimate": 0.37259642791302433
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05735161270315814,
-    "producer_p95_seconds": 0.18606425799953286,
-    "gpu_update_seconds": 0.9546017950736376,
-    "producer_batches_per_second": 17.43630131511426,
-    "gpu_batches_per_second": 1.047557217219417,
-    "producer_headroom": 16.6447245348529,
+    "producer_batch_seconds": 0.11849719262492897,
+    "producer_p95_seconds": 0.20257940599913127,
+    "gpu_update_seconds": 0.9518042346161931,
+    "producer_batches_per_second": 8.439018493587703,
+    "gpu_batches_per_second": 1.0506362165989327,
+    "producer_headroom": 8.032293538201143,
     "sustainable": true,
-    "buffer_seconds": 60.13991308963917,
+    "buffer_seconds": 60.91547101543636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309231.2768500555,
-    "upload_bytes_per_second": 4348227.509363677,
+    "generation_bytes_per_second": 329885.4610728684,
+    "upload_bytes_per_second": 4310254.658122835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.35383347631692,
+    "estimated_full_shards_per_hour": 70.78574060573138,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 10873151,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9900608629995986,
+      "bytes": 1198965,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -104,3 +104,4 @@
 {"step": 935400, "loss": 0.5121860504150391, "lr": 3e-05, "grad_norm": 0.3330061097951168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136916054698, "tokens_per_second": 152859.43604215118, "data_wait_s": 1.077599881682545e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935500, "loss": 0.5125136971473694, "lr": 3e-05, "grad_norm": 0.3212813415458014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136930735310, "tokens_per_second": 153590.0497284079, "data_wait_s": 1.3070999557385221e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935600, "loss": 0.537193238735199, "lr": 3e-05, "grad_norm": 0.3374356437921649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136945452772, "tokens_per_second": 153879.23339808328, "data_wait_s": 1.5105000784387812e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 935700, "loss": 0.5434572696685791, "lr": 3e-05, "grad_norm": 0.32923772548568175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136960174468, "tokens_per_second": 153736.71112189864, "data_wait_s": 1.3414000932243653e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:faff14b0e66d31b0c8913cdbcc1b55ca956c3b5f38bb3d175a5e9037812244ba
-size 226239
+oid sha256:3b7803da02388e1bb4f6239e93851c09d74d73cf65c714b95f9b368dac854771
+size 228357