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": 76671.34370399594,
-      "required_pages_per_second_upper_estimate": 0.005304301459469948
+      "required_token_positions_per_second": 77453.10452099967,
+      "required_pages_per_second_upper_estimate": 0.005358385486725273
     },
     {
       "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": 9583.917962999492,
-      "required_pages_per_second_upper_estimate": 0.004012586316142715
+      "required_token_positions_per_second": 9681.638065124958,
+      "required_pages_per_second_upper_estimate": 0.004053499682274867
     },
     {
       "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": 19167.835925998985,
-      "required_pages_per_second_upper_estimate": 0.003992154733559502
+      "required_token_positions_per_second": 19363.276130249917,
+      "required_pages_per_second_upper_estimate": 0.004032859774000186
     },
     {
       "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": 28751.75388899848,
-      "required_pages_per_second_upper_estimate": 0.001723623744259243
+      "required_token_positions_per_second": 29044.914195374873,
+      "required_pages_per_second_upper_estimate": 0.0017411982569966386
     },
     {
       "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": 28751.75388899848,
-      "required_pages_per_second_upper_estimate": 0.0022353223992695706
+      "required_token_positions_per_second": 29044.914195374873,
+      "required_pages_per_second_upper_estimate": 0.0022581143236144238
     },
     {
       "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": 28751.75388899848,
-      "required_pages_per_second_upper_estimate": 0.36879662765996435
+      "required_token_positions_per_second": 29044.914195374873,
+      "required_pages_per_second_upper_estimate": 0.37255697329914794
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09754042230423465,
-    "producer_p95_seconds": 0.34494055299728643,
-    "gpu_update_seconds": 0.961610902303222,
-    "producer_batches_per_second": 10.252159836677123,
-    "gpu_batches_per_second": 1.0399216539712992,
-    "producer_headroom": 9.858588671103941,
+    "producer_batch_seconds": 0.43779749998429907,
+    "producer_p95_seconds": 0.6190619110057014,
+    "gpu_update_seconds": 0.9519050328061455,
+    "producer_batches_per_second": 2.2841610562779904,
+    "gpu_batches_per_second": 1.0505249636637324,
+    "producer_headroom": 2.1743044052108202,
     "sustainable": true,
-    "buffer_seconds": 60.58148684510299,
+    "buffer_seconds": 60.92192209959331,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366223.02480693226,
-    "upload_bytes_per_second": 5080294.423461347,
+    "generation_bytes_per_second": 348183.8477276332,
+    "upload_bytes_per_second": 4993838.710405627,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.58293588787056,
+    "estimated_full_shards_per_hour": 74.7121484172034,
     "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": 1.675573243999679,
-      "bytes": 8344893,
-      "pages": 36,
-      "shards": 3
-    },
-    {
-      "seconds": 1.776490742999158,
-      "bytes": 11106825,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.2736427769996226,
       "bytes": 1040054,
@@ -562,6 +550,18 @@
       "bytes": 6280980,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9224069299962139,
+      "bytes": 730166,
+      "pages": 20,
+      "shards": 1
+    },
+    {
+      "seconds": 1.1284890140013886,
+      "bytes": 4366438,
+      "pages": 9,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -398,3 +398,4 @@
 {"step": 964800, "loss": 0.5158869028091431, "lr": 3e-05, "grad_norm": 0.33215481712055117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141220304902, "tokens_per_second": 152368.5144324505, "data_wait_s": 1.2736003554891795e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964900, "loss": 0.5310418605804443, "lr": 3e-05, "grad_norm": 0.3544124334635774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141234854891, "tokens_per_second": 151755.32677238516, "data_wait_s": 9.47499938774854e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 965000, "loss": 0.5895737409591675, "lr": 3e-05, "grad_norm": 0.3801122804407297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141249570216, "tokens_per_second": 153698.0028411782, "data_wait_s": 1.350299862679094e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 965100, "loss": 0.5409970879554749, "lr": 3e-05, "grad_norm": 0.34271652718062795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141264249572, "tokens_per_second": 131399.95456990355, "data_wait_s": 1.9553997844923288e-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 sha256:1a1b389048edf32fb0b311a07d31cd7c31c993a0a374540349d76c33cca9f1bf
-size 853629
+oid sha256:b22a16d7058bb76031998fa5e186eeb13a5d304876d7e8a79d55ced9695a6b3a
+size 855747