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": 76861.82107725846,
-      "required_pages_per_second_upper_estimate": 0.005317479126120647
+      "required_token_positions_per_second": 76649.58800367286,
+      "required_pages_per_second_upper_estimate": 0.0053027963496414185
     },
     {
       "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": 9607.727634657307,
-      "required_pages_per_second_upper_estimate": 0.0040225549284633585
+      "required_token_positions_per_second": 9581.198500459108,
+      "required_pages_per_second_upper_estimate": 0.004011447733965891
     },
     {
       "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": 19215.455269314614,
-      "required_pages_per_second_upper_estimate": 0.004002072586965615
+      "required_token_positions_per_second": 19162.397000918216,
+      "required_pages_per_second_upper_estimate": 0.003991021948899277
     },
     {
       "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": 28823.182903971923,
-      "required_pages_per_second_upper_estimate": 0.0017279058046411103
+      "required_token_positions_per_second": 28743.595501377324,
+      "required_pages_per_second_upper_estimate": 0.0017231346613784913
     },
     {
       "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": 28823.182903971923,
-      "required_pages_per_second_upper_estimate": 0.002240875691000723
+      "required_token_positions_per_second": 28743.595501377324,
+      "required_pages_per_second_upper_estimate": 0.0022346881205169804
     },
     {
       "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": 28823.182903971923,
-      "required_pages_per_second_upper_estimate": 0.36971284236954277
+      "required_token_positions_per_second": 28743.595501377324,
+      "required_pages_per_second_upper_estimate": 0.3686919806233543
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11450032085159023,
-    "producer_p95_seconds": 0.2987335059988254,
-    "gpu_update_seconds": 0.9592278580791305,
-    "producer_batches_per_second": 8.733599980878234,
-    "gpu_batches_per_second": 1.0425051686911142,
-    "producer_headroom": 8.377512402977763,
+    "producer_batch_seconds": 0.2838415886249379,
+    "producer_p95_seconds": 0.3089924330015492,
+    "gpu_update_seconds": 0.9618838394339071,
+    "producer_batches_per_second": 3.5230918937724036,
+    "gpu_batches_per_second": 1.0396265734005108,
+    "producer_headroom": 3.388805157460274,
     "sustainable": true,
-    "buffer_seconds": 60.43135505898522,
+    "buffer_seconds": 60.598681884336145,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 163,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 521152.36748440267,
+    "generation_bytes_per_second": 595465.3045448787,
     "upload_bytes_per_second": 4795947.702091147,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.82716625594196,
+    "estimated_full_shards_per_hour": 63.88649631624112,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -318,3 +318,4 @@
 {"step": 956800, "loss": 0.5233384370803833, "lr": 3e-05, "grad_norm": 0.33955261842588325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140049018692, "tokens_per_second": 152160.04100585048, "data_wait_s": 1.3610999303637072e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 956900, "loss": 0.524031937122345, "lr": 3e-05, "grad_norm": 0.3326202490118902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140063773378, "tokens_per_second": 153690.96492832043, "data_wait_s": 1.3233002391643822e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 957000, "loss": 0.5587459206581116, "lr": 3e-05, "grad_norm": 0.3490230127132062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140078464846, "tokens_per_second": 152948.65334969288, "data_wait_s": 1.628300014999695e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 957100, "loss": 0.46292242407798767, "lr": 3e-05, "grad_norm": 0.3678985811875492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140093208546, "tokens_per_second": 153248.6847400724, "data_wait_s": 1.1796997569035739e-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 sha256:352afcd8981b042a6ffa12249df1fc3f4f3757374dfaf5fdbb2505b6122d2d05
-size 682623
+oid sha256:12d85f64b8ffd3017aa94856794f9bdaa9119be9ca92275c1002b2dd45c67b4a
+size 684741