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": 77398.93500300075,
-      "required_pages_per_second_upper_estimate": 0.005354637913779509
+      "required_token_positions_per_second": 76722.54344231935,
+      "required_pages_per_second_upper_estimate": 0.005307843576166923
     },
     {
       "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": 9674.866875375094,
-      "required_pages_per_second_upper_estimate": 0.004050664726525119
+      "required_token_positions_per_second": 9590.317930289919,
+      "required_pages_per_second_upper_estimate": 0.004015265848800701
     },
     {
       "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": 19349.733750750187,
-      "required_pages_per_second_upper_estimate": 0.004030039253486956
+      "required_token_positions_per_second": 19180.635860579838,
+      "required_pages_per_second_upper_estimate": 0.003994820622375718
     },
     {
       "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": 29024.600626125284,
-      "required_pages_per_second_upper_estimate": 0.00173998048953715
+      "required_token_positions_per_second": 28770.953790869757,
+      "required_pages_per_second_upper_estimate": 0.0017247747490598735
     },
     {
       "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": 29024.600626125284,
-      "required_pages_per_second_upper_estimate": 0.0022565350329552163
+      "required_token_positions_per_second": 28770.953790869757,
+      "required_pages_per_second_upper_estimate": 0.002236815107188619
     },
     {
       "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": 29024.600626125284,
-      "required_pages_per_second_upper_estimate": 0.37229641264382557
+      "required_token_positions_per_second": 28770.953790869757,
+      "required_pages_per_second_upper_estimate": 0.3690429033859206
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.21329784839062427,
-    "producer_p95_seconds": 0.21022852600071928,
-    "gpu_update_seconds": 0.9525712465829351,
-    "producer_batches_per_second": 4.688279828161436,
-    "gpu_batches_per_second": 1.049790242553721,
-    "producer_headroom": 4.4659205602413685,
+    "producer_batch_seconds": 0.42486412724210254,
+    "producer_p95_seconds": 0.20935687199926178,
+    "gpu_update_seconds": 0.9609691844409372,
+    "producer_batches_per_second": 2.353693653759958,
+    "gpu_batches_per_second": 1.0406160948665277,
+    "producer_headroom": 2.261827070877516,
     "sustainable": true,
-    "buffer_seconds": 60.011988534724914,
+    "buffer_seconds": 60.541058619779044,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430464.99988839874,
+    "generation_bytes_per_second": 380174.84375796467,
     "upload_bytes_per_second": 5573126.438450026,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.36776826371165,
+    "estimated_full_shards_per_hour": 81.57667145303921,
     "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": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 568000, "loss": 0.5760231018066406, "lr": 0.0003, "grad_norm": 0.14936431707598977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83138360447, "tokens_per_second": 151341.9966538106, "data_wait_s": 1.1370999345672317e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568100, "loss": 0.5821261405944824, "lr": 0.0003, "grad_norm": 0.19712639032358825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83152926119, "tokens_per_second": 151624.50268147435, "data_wait_s": 1.1381998774595559e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568200, "loss": 0.5839268565177917, "lr": 0.0003, "grad_norm": 0.16720505818914366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83167589476, "tokens_per_second": 152423.64082107524, "data_wait_s": 1.2625001545529813e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 568300, "loss": 0.5627555847167969, "lr": 0.0003, "grad_norm": 0.18035638580757987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83182231142, "tokens_per_second": 152265.45215841712, "data_wait_s": 1.2732998584397137e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62476b759dddc8d89901d5fb05b488c23da91c9a43bf816746a532b2a4a3f182
-size 236829
+oid sha256:5c1c90d519c4f095af222d10692a959d699fe2b8f10864797524ff9cf530e70d
+size 238947