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": 76707.55563960897,
-      "required_pages_per_second_upper_estimate": 0.005306806685198924
+      "required_token_positions_per_second": 76535.69251029009,
+      "required_pages_per_second_upper_estimate": 0.0052949167951352434
     },
     {
       "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": 9588.44445495112,
-      "required_pages_per_second_upper_estimate": 0.0040144814637989605
+      "required_token_positions_per_second": 9566.96156378626,
+      "required_pages_per_second_upper_estimate": 0.004005487025882015
     },
     {
       "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": 19176.88890990224,
-      "required_pages_per_second_upper_estimate": 0.0039940402313633095
+      "required_token_positions_per_second": 19133.92312757252,
+      "required_pages_per_second_upper_estimate": 0.003985091591987905
     },
     {
       "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": 28765.333364853363,
-      "required_pages_per_second_upper_estimate": 0.0017244378131020842
+      "required_token_positions_per_second": 28700.88469135878,
+      "required_pages_per_second_upper_estimate": 0.0017205742135335104
     },
     {
       "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": 28765.333364853363,
-      "required_pages_per_second_upper_estimate": 0.002236378143787486
+      "required_token_positions_per_second": 28700.88469135878,
+      "required_pages_per_second_upper_estimate": 0.002231367542903038
     },
     {
       "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": 28765.333364853363,
-      "required_pages_per_second_upper_estimate": 0.36897081059572556
+      "required_token_positions_per_second": 28700.88469135878,
+      "required_pages_per_second_upper_estimate": 0.36814413221173126
     }
   ],
   "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": 0.08560508857823379,
-    "producer_p95_seconds": 0.2776440259985975,
-    "gpu_update_seconds": 0.9611569471251612,
-    "producer_batches_per_second": 11.681548569231468,
-    "gpu_batches_per_second": 1.0404128097820227,
-    "producer_headroom": 11.227801560496813,
+    "producer_batch_seconds": 0.08155562882814138,
+    "producer_p95_seconds": 0.24468312099998002,
+    "gpu_update_seconds": 0.9633152530773457,
+    "producer_batches_per_second": 12.2615693652152,
+    "gpu_batches_per_second": 1.038081766903891,
+    "producer_headroom": 11.81175679617771,
     "sustainable": true,
-    "buffer_seconds": 60.552887668885155,
+    "buffer_seconds": 60.688860943872776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 225461.53839994952,
+    "generation_bytes_per_second": 234676.9178698362,
     "upload_bytes_per_second": 3989924.833774804,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.378797664631506,
+    "estimated_full_shards_per_hour": 50.35620357581439,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 820400, "loss": 0.49906840920448303, "lr": 3e-05, "grad_norm": 0.2801640626981123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120084012610, "tokens_per_second": 118963.99736858156, "data_wait_s": 1.4907000149833038e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 820500, "loss": 0.5685487985610962, "lr": 3e-05, "grad_norm": 0.292315157138596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120098478571, "tokens_per_second": 149555.9637116822, "data_wait_s": 1.030999919748865e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 820600, "loss": 0.4981805384159088, "lr": 3e-05, "grad_norm": 0.268580052436513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120112987406, "tokens_per_second": 150346.26286204386, "data_wait_s": 1.4757999451830983e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 820700, "loss": 0.5129390954971313, "lr": 3e-05, "grad_norm": 0.2830667316656344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120127553733, "tokens_per_second": 149909.6327589666, "data_wait_s": 1.0944000678136945e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d690284561396adea2bad7643f0a9ac85cfaca8e9c02635ff39e4bac52c73499
-size 439605
+oid sha256:4379ce695dcc0bb76cb0791f68bade271a919c591bc6dd05de35490afbad0a19
+size 441723