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": 76467.45974408176,
-      "required_pages_per_second_upper_estimate": 0.005290196294047121
+      "required_token_positions_per_second": 76895.03207795933,
+      "required_pages_per_second_upper_estimate": 0.0053197767402092715
     },
     {
       "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": 9558.43246801022,
-      "required_pages_per_second_upper_estimate": 0.00400191607158836
+      "required_token_positions_per_second": 9611.879009744916,
+      "required_pages_per_second_upper_estimate": 0.00402429302252197
     },
     {
       "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": 19116.86493602044,
-      "required_pages_per_second_upper_estimate": 0.00398153882054237
+      "required_token_positions_per_second": 19223.758019489833,
+      "required_pages_per_second_upper_estimate": 0.004003801830868866
     },
     {
       "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": 28675.29740403066,
-      "required_pages_per_second_upper_estimate": 0.0017190402947277153
+      "required_token_positions_per_second": 28835.63702923475,
+      "required_pages_per_second_upper_estimate": 0.0017286524104342684
     },
     {
       "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": 28675.29740403066,
-      "required_pages_per_second_upper_estimate": 0.002229378243859859
+      "required_token_positions_per_second": 28835.63702923475,
+      "required_pages_per_second_upper_estimate": 0.00224184394446
     },
     {
       "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": 28675.29740403066,
-      "required_pages_per_second_upper_estimate": 0.3678159259633748
+      "required_token_positions_per_second": 28835.63702923475,
+      "required_pages_per_second_upper_estimate": 0.3698725905162164
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23820933434370772,
-    "producer_p95_seconds": 0.3158549279978615,
-    "gpu_update_seconds": 0.9641748300093912,
-    "producer_batches_per_second": 4.197988306189207,
-    "gpu_batches_per_second": 1.037156300782359,
-    "producer_headroom": 4.047594661501391,
+    "producer_batch_seconds": 0.38885338139061787,
+    "producer_p95_seconds": 0.32565086799877463,
+    "gpu_update_seconds": 0.9588135671138226,
+    "producer_batches_per_second": 2.571663377141788,
+    "gpu_batches_per_second": 1.0429556217171134,
+    "producer_headroom": 2.4657457360532975,
     "sustainable": true,
-    "buffer_seconds": 60.743014290591645,
+    "buffer_seconds": 60.405254728170824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 498510.1371353484,
-    "upload_bytes_per_second": 4784462.0400673,
+    "generation_bytes_per_second": 447110.93887465657,
+    "upload_bytes_per_second": 4795297.797758385,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.96867070718135,
+    "estimated_full_shards_per_hour": 95.93959927253506,
     "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": 1.1428929149988107,
-      "bytes": 5656732,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3837304669978039,
       "bytes": 7550021,
@@ -562,6 +556,12 @@
       "bytes": 10634260,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.4519578809995437,
+      "bytes": 8113266,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -341,3 +341,4 @@
 {"step": 959100, "loss": 0.5026146173477173, "lr": 3e-05, "grad_norm": 0.3385202518930032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140386033764, "tokens_per_second": 153723.02122442852, "data_wait_s": 1.0247997124679387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959200, "loss": 0.5705493092536926, "lr": 3e-05, "grad_norm": 0.3536448166679467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140400686195, "tokens_per_second": 152233.72911656197, "data_wait_s": 1.6534002497792244e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959300, "loss": 0.5473478436470032, "lr": 3e-05, "grad_norm": 0.3774777091774848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140415402441, "tokens_per_second": 152828.73968524157, "data_wait_s": 1.2432996300049126e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 959400, "loss": 0.5365502238273621, "lr": 3e-05, "grad_norm": 0.347566418884559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140430128981, "tokens_per_second": 152366.0358528447, "data_wait_s": 1.1727002856787294e-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 sha256:0f2772a6435011e411e8e137f6b64119893e93204935bb813518dc61b9124f4e
-size 731337
+oid sha256:60f31834c6c71674656ff788e914976cbf41560d642e2f862127a581946d21ba
+size 733455