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": 76492.77939521884,
-      "required_pages_per_second_upper_estimate": 0.005291947966262469
+      "required_token_positions_per_second": 76354.41069708552,
+      "required_pages_per_second_upper_estimate": 0.005282375298665994
     },
     {
       "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": 9561.597424402355,
-      "required_pages_per_second_upper_estimate": 0.004003241172737941
+      "required_token_positions_per_second": 9544.30133713569,
+      "required_pages_per_second_upper_estimate": 0.003995999662182763
     },
     {
       "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": 19123.19484880471,
-      "required_pages_per_second_upper_estimate": 0.003982857174444307
+      "required_token_positions_per_second": 19088.60267427138,
+      "required_pages_per_second_upper_estimate": 0.003975652536746004
     },
     {
       "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": 28684.792273207062,
-      "required_pages_per_second_upper_estimate": 0.0017196094976370147
+      "required_token_positions_per_second": 28632.904011407074,
+      "required_pages_per_second_upper_estimate": 0.0017164988755708937
     },
     {
       "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": 28684.792273207062,
-      "required_pages_per_second_upper_estimate": 0.002230116428174809
+      "required_token_positions_per_second": 28632.904011407074,
+      "required_pages_per_second_upper_estimate": 0.002226082344052203
     },
     {
       "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": 28684.792273207062,
-      "required_pages_per_second_upper_estimate": 0.3679377159503734
+      "required_token_positions_per_second": 28632.904011407074,
+      "required_pages_per_second_upper_estimate": 0.367272149041278
     }
   ],
   "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": 0.17687175094539498,
-    "producer_p95_seconds": 0.2654227519997221,
-    "gpu_update_seconds": 0.9638556813194887,
-    "producer_batches_per_second": 5.653814103467131,
-    "gpu_batches_per_second": 1.0374997205297694,
-    "producer_headroom": 5.449460844751046,
+    "producer_batch_seconds": 0.09542631714850813,
+    "producer_p95_seconds": 0.27993752700058394,
+    "gpu_update_seconds": 0.9656023709290997,
+    "producer_batches_per_second": 10.47928946522939,
+    "gpu_batches_per_second": 1.0356229749496193,
+    "producer_headroom": 10.118826753277837,
     "sustainable": true,
-    "buffer_seconds": 60.72290792312779,
+    "buffer_seconds": 60.832949368533285,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 520277.7836380742,
-    "upload_bytes_per_second": 5141985.905198482,
+    "generation_bytes_per_second": 525196.9752932341,
+    "upload_bytes_per_second": 5170694.310309985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.63950092178983,
+    "estimated_full_shards_per_hour": 112.69504493806618,
     "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": 1.6974678700007644,
-      "bytes": 6266676,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.6492898739998054,
       "bytes": 1094803,
@@ -562,6 +556,12 @@
       "bytes": 16881065,
       "pages": 36,
       "shards": 5
+    },
+    {
+      "seconds": 1.1940166240019607,
+      "bytes": 6068573,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 741600, "loss": 0.5574122071266174, "lr": 5e-05, "grad_norm": 0.23297379460876017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108550071163, "tokens_per_second": 150425.03777298765, "data_wait_s": 1.635100124985911e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 741700, "loss": 0.5075755715370178, "lr": 5e-05, "grad_norm": 0.22462851633801031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108564764266, "tokens_per_second": 151220.2962149451, "data_wait_s": 1.2710999726550654e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 741800, "loss": 0.5595993995666504, "lr": 5e-05, "grad_norm": 0.2460738789524235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108579386178, "tokens_per_second": 150463.40072265756, "data_wait_s": 1.593300112290308e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 741900, "loss": 0.549426257610321, "lr": 5e-05, "grad_norm": 0.23333403917219317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108593877800, "tokens_per_second": 148868.92985597448, "data_wait_s": 1.5596000594086945e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4615b6db54668a77cb909c924dcce213e20d7ba212472923bc00ffce02a0d966
-size 465021
+oid sha256:36e4d8050166e8378f9ea79df0d6ab6989acddd3b7671423debe260679674118
+size 467139