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": 76818.30058864875,
-      "required_pages_per_second_upper_estimate": 0.0053144682777371815
+      "required_token_positions_per_second": 76772.56937006339,
+      "required_pages_per_second_upper_estimate": 0.005311304485924399
     },
     {
       "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": 9602.287573581094,
-      "required_pages_per_second_upper_estimate": 0.004020277288492142
+      "required_token_positions_per_second": 9596.571171257923,
+      "required_pages_per_second_upper_estimate": 0.004017883950211485
     },
     {
       "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": 19204.575147162188,
-      "required_pages_per_second_upper_estimate": 0.00399980654444938
+      "required_token_positions_per_second": 19193.142342515846,
+      "required_pages_per_second_upper_estimate": 0.0039974253927448815
     },
     {
       "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": 28806.86272074328,
-      "required_pages_per_second_upper_estimate": 0.00172692743457603
+      "required_token_positions_per_second": 28789.713513773768,
+      "required_pages_per_second_upper_estimate": 0.0017258993658035306
     },
     {
       "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": 28806.86272074328,
-      "required_pages_per_second_upper_estimate": 0.0022396068685395956
+      "required_token_positions_per_second": 28789.713513773768,
+      "required_pages_per_second_upper_estimate": 0.0022382735931291056
     },
     {
       "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": 28806.86272074328,
-      "required_pages_per_second_upper_estimate": 0.3695035045823332
+      "required_token_positions_per_second": 28789.713513773768,
+      "required_pages_per_second_upper_estimate": 0.36928353296871214
     }
   ],
   "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": 0.2141030928124934,
-    "producer_p95_seconds": 0.17224195699964184,
-    "gpu_update_seconds": 0.9597712971392471,
-    "producer_batches_per_second": 4.670647148828332,
-    "gpu_batches_per_second": 1.041914884286143,
-    "producer_headroom": 4.482753072510694,
+    "producer_batch_seconds": 0.099020675843704,
+    "producer_p95_seconds": 0.17838320000009844,
+    "gpu_update_seconds": 0.9603430053853248,
+    "producer_batches_per_second": 10.098900976786078,
+    "gpu_batches_per_second": 1.0412946149368405,
+    "producer_headroom": 9.698408915135536,
     "sustainable": true,
-    "buffer_seconds": 60.46559171977257,
+    "buffer_seconds": 60.50160933927546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491018.1034788689,
-    "upload_bytes_per_second": 5468519.706539595,
+    "generation_bytes_per_second": 489227.7107253315,
+    "upload_bytes_per_second": 5501616.782450729,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.36105469011831,
+    "estimated_full_shards_per_hour": 104.97687808341941,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 9980497,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1220608109997556,
+      "bytes": 7908917,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -52,3 +52,4 @@
 {"step": 609600, "loss": 0.5150657892227173, "lr": 0.0001, "grad_norm": 0.1393956148179828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89227678418, "tokens_per_second": 154392.57602208518, "data_wait_s": 1.1154999810969457e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609700, "loss": 0.6639874577522278, "lr": 0.0001, "grad_norm": 0.17147602627155376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89242403577, "tokens_per_second": 152148.1796035995, "data_wait_s": 1.30259995785309e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609800, "loss": 0.5783213973045349, "lr": 0.0001, "grad_norm": 0.16062904182535676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89257200749, "tokens_per_second": 153469.19244595084, "data_wait_s": 1.6522999430890195e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 609900, "loss": 0.5257395505905151, "lr": 0.0001, "grad_norm": 0.13698058650701322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89271990677, "tokens_per_second": 153815.20690466673, "data_wait_s": 1.3779000255453866e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ffe9fb867b1611397b21df5d46daffc33fae50b4935baaf5388d7c55003f369
-size 115320
+oid sha256:7d6939036e0a4b3b2001083eef7ffb6c005dd6d17517eee3bb4c7c6ff1431a4e
+size 117438