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": 76925.50506245835,
-      "required_pages_per_second_upper_estimate": 0.005321884931983979
+      "required_token_positions_per_second": 77508.45942556956,
+      "required_pages_per_second_upper_estimate": 0.005362215067464492
     },
     {
       "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": 9615.688132807294,
-      "required_pages_per_second_upper_estimate": 0.004025887822804654
+      "required_token_positions_per_second": 9688.557428196194,
+      "required_pages_per_second_upper_estimate": 0.004056396675100062
     },
     {
       "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": 19231.37626561459,
-      "required_pages_per_second_upper_estimate": 0.004005388510629992
+      "required_token_positions_per_second": 19377.11485639239,
+      "required_pages_per_second_upper_estimate": 0.004035742015703912
     },
     {
       "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": 28847.06439842188,
-      "required_pages_per_second_upper_estimate": 0.0017293374637684545
+      "required_token_positions_per_second": 29065.672284588585,
+      "required_pages_per_second_upper_estimate": 0.0017424426727492338
     },
     {
       "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": 28847.06439842188,
-      "required_pages_per_second_upper_estimate": 0.0022427323721506145
+      "required_token_positions_per_second": 29065.672284588585,
+      "required_pages_per_second_upper_estimate": 0.0022597281737456046
     },
     {
       "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": 28847.06439842188,
-      "required_pages_per_second_upper_estimate": 0.3700191685383959
+      "required_token_positions_per_second": 29065.672284588585,
+      "required_pages_per_second_upper_estimate": 0.3728232357792818
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.15425020999205685,
-    "producer_p95_seconds": 0.21618036200015922,
-    "gpu_update_seconds": 0.9584337462605909,
-    "producer_batches_per_second": 6.48297334604274,
-    "gpu_batches_per_second": 1.0433689380216247,
-    "producer_headroom": 6.213500430955301,
+    "producer_batch_seconds": 0.061289952539112846,
+    "producer_p95_seconds": 0.1976377289993252,
+    "gpu_update_seconds": 0.9512252023380765,
+    "producer_batches_per_second": 16.315887981179937,
+    "gpu_batches_per_second": 1.0512757626080942,
+    "producer_headroom": 15.520083846223276,
     "sustainable": true,
-    "buffer_seconds": 60.381326014417226,
+    "buffer_seconds": 60.878412949636896,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440046.26260620257,
+    "generation_bytes_per_second": 441340.0148192776,
     "upload_bytes_per_second": 5786923.577862919,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.42368420257147,
+    "estimated_full_shards_per_hour": 94.70129331048724,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 494300, "loss": 0.6222625970840454, "lr": 0.0003, "grad_norm": 0.16598529522696565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72349821711, "tokens_per_second": 152517.2418838022, "data_wait_s": 9.6160001703538e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 494400, "loss": 0.6011156439781189, "lr": 0.0003, "grad_norm": 0.162331143455663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72364325594, "tokens_per_second": 151034.67332909734, "data_wait_s": 2.25710009544855e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 494500, "loss": 0.5836164355278015, "lr": 0.0003, "grad_norm": 0.18662898146005005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72378885512, "tokens_per_second": 152257.305106254, "data_wait_s": 7.495000318158418e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 494600, "loss": 0.5923532247543335, "lr": 0.0003, "grad_norm": 0.15387534114428109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72393405116, "tokens_per_second": 151688.3820863191, "data_wait_s": 1.3479999324772507e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8cf5e0d1364194737c13c1ef6949ab4a1c299029ae0f8d77ec2a0eb33a9f0eea
-size 202158
+oid sha256:00cb81ee7a7eaf90d068c074f5ffb33385af22ac009071b4e3697525072dde65
+size 204276