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": 77671.27318388631,
-      "required_pages_per_second_upper_estimate": 0.005373478901044813
+      "required_token_positions_per_second": 77248.9298521678,
+      "required_pages_per_second_upper_estimate": 0.005344260209384983
     },
     {
       "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": 9708.909147985789,
-      "required_pages_per_second_upper_estimate": 0.0040649175151837285
+      "required_token_positions_per_second": 9656.116231520975,
+      "required_pages_per_second_upper_estimate": 0.00404281422350137
     },
     {
       "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": 19417.818295971578,
-      "required_pages_per_second_upper_estimate": 0.0040442194687463715
+      "required_token_positions_per_second": 19312.23246304195,
+      "required_pages_per_second_upper_estimate": 0.004022228724232793
     },
     {
       "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": 29126.72744395737,
-      "required_pages_per_second_upper_estimate": 0.001746102836327562
+      "required_token_positions_per_second": 28968.348694562923,
+      "required_pages_per_second_upper_estimate": 0.0017366082721317127
     },
     {
       "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": 29126.72744395737,
-      "required_pages_per_second_upper_estimate": 0.002264474943833263
+      "required_token_positions_per_second": 28968.348694562923,
+      "required_pages_per_second_upper_estimate": 0.0022521616927023407
     },
     {
       "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": 29126.72744395737,
-      "required_pages_per_second_upper_estimate": 0.37360638580774197
+      "required_token_positions_per_second": 28968.348694562923,
+      "required_pages_per_second_upper_estimate": 0.37157487326436195
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 0.046032591781246524,
-    "producer_p95_seconds": 0.12232808299995668,
-    "gpu_update_seconds": 0.9492312534320039,
-    "producer_batches_per_second": 21.72373879689728,
-    "gpu_batches_per_second": 1.0534840655366526,
-    "producer_headroom": 20.620851807408258,
+    "producer_batch_seconds": 0.04515125417969479,
+    "producer_p95_seconds": 0.12357050199989317,
+    "gpu_update_seconds": 0.9544209886285048,
+    "producer_batches_per_second": 22.147779018942853,
+    "gpu_batches_per_second": 1.0477556674827446,
+    "producer_headroom": 21.138305147185093,
     "sustainable": true,
-    "buffer_seconds": 60.75080021964825,
+    "buffer_seconds": 60.1285222835958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 56,
+    "upload_pages_per_commit": 61,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -5,3 +5,4 @@
 {"step": 645500, "loss": 0.5504806041717529, "lr": 0.0001, "grad_norm": 0.1736480488849375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94484547368, "tokens_per_second": 152933.4877733488, "data_wait_s": 9.080000040739833e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 645600, "loss": 0.54863041639328, "lr": 0.0001, "grad_norm": 0.18142711624629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94499089731, "tokens_per_second": 151655.98749001828, "data_wait_s": 9.263000038117752e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 645700, "loss": 0.557036817073822, "lr": 0.0001, "grad_norm": 0.175969469020709, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94513673484, "tokens_per_second": 152917.05001293845, "data_wait_s": 1.1775999837482232e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 645800, "loss": 0.518687903881073, "lr": 0.0001, "grad_norm": 0.17502375398215597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94528455980, "tokens_per_second": 154864.7007960821, "data_wait_s": 1.1122000159957679e-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 sha256:14f309757aa5132dda13c1cab28b95a6a19dc5557aa19fcc375bd676f1338d90
-size 14991
+oid sha256:43641ea9bc21cf1e04b582299808eebe67aeeda6268bbf48aca823164c39424b
+size 17109