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": 76814.43520626044,
-      "required_pages_per_second_upper_estimate": 0.005314200861614635
+      "required_token_positions_per_second": 76955.31394533147,
+      "required_pages_per_second_upper_estimate": 0.005323947179667279
     },
     {
       "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": 9601.804400782556,
-      "required_pages_per_second_upper_estimate": 0.004020074994131189
+      "required_token_positions_per_second": 9619.414243166433,
+      "required_pages_per_second_upper_estimate": 0.004027447867401993
     },
     {
       "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": 19203.60880156511,
-      "required_pages_per_second_upper_estimate": 0.003999605280145756
+      "required_token_positions_per_second": 19238.828486332866,
+      "required_pages_per_second_upper_estimate": 0.0040069406116773326
     },
     {
       "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": 28805.41320234767,
-      "required_pages_per_second_upper_estimate": 0.0017268405381614996
+      "required_token_positions_per_second": 28858.242729499296,
+      "required_pages_per_second_upper_estimate": 0.0017300075876482224
     },
     {
       "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": 28805.41320234767,
-      "required_pages_per_second_upper_estimate": 0.0022394941748600946
+      "required_token_positions_per_second": 28858.242729499296,
+      "required_pages_per_second_upper_estimate": 0.0022436014382236015
     },
     {
       "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": 28805.41320234767,
-      "required_pages_per_second_upper_estimate": 0.36948491171672593
+      "required_token_positions_per_second": 28858.242729499296,
+      "required_pages_per_second_upper_estimate": 0.37016255216710015
     }
   ],
   "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": 0.08977833807031743,
-    "producer_p95_seconds": 0.2332394889999705,
-    "gpu_update_seconds": 0.959819593830602,
-    "producer_batches_per_second": 11.138544347042448,
-    "gpu_batches_per_second": 1.041862456682135,
-    "producer_headroom": 10.690993111042431,
+    "producer_batch_seconds": 0.08477259781227531,
+    "producer_p95_seconds": 0.2500173309999809,
+    "gpu_update_seconds": 0.9580624939348032,
+    "producer_batches_per_second": 11.796264663428744,
+    "gpu_batches_per_second": 1.0437732468713576,
+    "producer_headroom": 11.301558742559534,
     "sustainable": true,
-    "buffer_seconds": 60.468634411327926,
+    "buffer_seconds": 60.357937117892604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306135.2958593283,
-    "upload_bytes_per_second": 5071152.276700092,
+    "generation_bytes_per_second": 330331.7014523559,
+    "upload_bytes_per_second": 5060966.021529688,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.68950802645575,
+    "estimated_full_shards_per_hour": 70.88149340322502,
     "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": 1.1447629150006833,
-      "bytes": 7321171,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0708932610014017,
       "bytes": 6812845,
@@ -562,6 +556,12 @@
       "bytes": 7549149,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1122719240011065,
+      "bytes": 6287535,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -192,3 +192,4 @@
 {"step": 779200, "loss": 0.5231711268424988, "lr": 3e-05, "grad_norm": 0.2526803621215525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114052935439, "tokens_per_second": 151390.63273724014, "data_wait_s": 1.5181001799646765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779300, "loss": 0.5392187833786011, "lr": 3e-05, "grad_norm": 0.2561384988112868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114067641327, "tokens_per_second": 153041.10515992413, "data_wait_s": 1.3837998267263174e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779400, "loss": 0.5333099365234375, "lr": 3e-05, "grad_norm": 0.24759982815345763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114082340629, "tokens_per_second": 153000.9783801805, "data_wait_s": 1.1845000699395314e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 779500, "loss": 0.5458994507789612, "lr": 3e-05, "grad_norm": 0.25431371027240035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114097047090, "tokens_per_second": 145218.2433116972, "data_wait_s": 1.3504002708941698e-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 sha256:7b45c02ba8c553416d2b7c1f04ed135b200df42a6c9ea862b47ea065e2ace1cc
-size 413406
+oid sha256:e3c5d451738de547eb9651f08426fc348bec9cf00b74517ed6ba14404cc5e87f
+size 415524