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": 75748.5152379929,
-      "required_pages_per_second_upper_estimate": 0.00524045804493481
+      "required_token_positions_per_second": 75598.02752697378,
+      "required_pages_per_second_upper_estimate": 0.0052300469559069095
     },
     {
       "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": 9468.564404749113,
-      "required_pages_per_second_upper_estimate": 0.003964290190159497
+      "required_token_positions_per_second": 9449.753440871722,
+      "required_pages_per_second_upper_estimate": 0.003956414432401628
     },
     {
       "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": 18937.128809498226,
-      "required_pages_per_second_upper_estimate": 0.003944104525348198
+      "required_token_positions_per_second": 18899.506881743444,
+      "required_pages_per_second_upper_estimate": 0.003936268869953831
     },
     {
       "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": 28405.693214247338,
-      "required_pages_per_second_upper_estimate": 0.0017028779352119669
+      "required_token_positions_per_second": 28349.260322615166,
+      "required_pages_per_second_upper_estimate": 0.0016994948695266544
     },
     {
       "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": 28405.693214247338,
-      "required_pages_per_second_upper_estimate": 0.002208417703967711
+      "required_token_positions_per_second": 28349.260322615166,
+      "required_pages_per_second_upper_estimate": 0.002204030294865367
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28405.693214247338,
-      "required_pages_per_second_upper_estimate": 0.36435773289525963
+      "required_token_positions_per_second": 28349.260322615166,
+      "required_pages_per_second_upper_estimate": 0.3636338723543203
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05537231212503002,
-    "producer_p95_seconds": 0.1873168469992379,
-    "gpu_update_seconds": 0.9733260086795803,
-    "producer_batches_per_second": 18.05956734734161,
-    "gpu_batches_per_second": 1.027404991834756,
-    "producer_headroom": 17.577846604668082,
+    "producer_batch_seconds": 0.8275764796875507,
+    "producer_p95_seconds": 0.2345323079989612,
+    "gpu_update_seconds": 0.9752635407543333,
+    "producer_batches_per_second": 1.2083475359009084,
+    "gpu_batches_per_second": 1.0253638716223656,
+    "producer_headroom": 1.1784572963244937,
     "sustainable": true,
-    "buffer_seconds": 60.346212538133976,
+    "buffer_seconds": 60.466339526768664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 89,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300493.16611940745,
-    "upload_bytes_per_second": 5386211.437909366,
+    "generation_bytes_per_second": 258854.082937507,
+    "upload_bytes_per_second": 5456007.611965329,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.47883832632702,
+    "estimated_full_shards_per_hour": 55.54406038373859,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 1235188,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2133236099998612,
+      "bytes": 11087663,
+      "pages": 9,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -99,3 +99,4 @@
 {"step": 769900, "loss": 0.5111449360847473, "lr": 3e-05, "grad_norm": 0.24221454782284982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112690918043, "tokens_per_second": 154322.27194159306, "data_wait_s": 8.94799995876383e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 770000, "loss": 0.5791223049163818, "lr": 3e-05, "grad_norm": 0.28168012182084734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112705486715, "tokens_per_second": 150574.57346352158, "data_wait_s": 1.4241999451769516e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770100, "loss": 0.5264490842819214, "lr": 3e-05, "grad_norm": 0.24843234261534192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112720043214, "tokens_per_second": 102502.96052938142, "data_wait_s": 1.5718000213382766e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 770200, "loss": 0.541844367980957, "lr": 3e-05, "grad_norm": 0.2595089970675744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112734606935, "tokens_per_second": 151611.935820005, "data_wait_s": 1.6105001122923568e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b060482a511d4fac153520fcbb7eceb25d41f5ce95dc534fce13a758b134603e
-size 215649
+oid sha256:c1d4c553951d46b4094927d880bafa39b2169d5fe02c3d63a2fe79f8aa8066f5
+size 217767