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": 76569.23155556619,
-      "required_pages_per_second_upper_estimate": 0.005297237104108751
+      "required_token_positions_per_second": 75465.33859747375,
+      "required_pages_per_second_upper_estimate": 0.005220867227883363
     },
     {
       "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": 9571.153944445774,
-      "required_pages_per_second_upper_estimate": 0.00400724228811729
+      "required_token_positions_per_second": 9433.167324684218,
+      "required_pages_per_second_upper_estimate": 0.003949470171911412
     },
     {
       "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": 19142.307888891548,
-      "required_pages_per_second_upper_estimate": 0.003986837916649633
+      "required_token_positions_per_second": 18866.334649368437,
+      "required_pages_per_second_upper_estimate": 0.003929359968760713
     },
     {
       "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": 28713.461833337322,
-      "required_pages_per_second_upper_estimate": 0.0017213281939909401
+      "required_token_positions_per_second": 28299.501974052655,
+      "required_pages_per_second_upper_estimate": 0.001696511932506931
     },
     {
       "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": 28713.461833337322,
-      "required_pages_per_second_upper_estimate": 0.0022323453603708683
+      "required_token_positions_per_second": 28299.501974052655,
+      "required_pages_per_second_upper_estimate": 0.002200161802128477
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28713.461833337322,
-      "required_pages_per_second_upper_estimate": 0.3683054582847491
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28299.501974052655,
+      "required_pages_per_second_upper_estimate": 0.3629956256853126
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06930161364063991,
-    "producer_p95_seconds": 0.1801968849995319,
-    "gpu_update_seconds": 0.9628932993338934,
-    "producer_batches_per_second": 14.429678437005096,
-    "gpu_batches_per_second": 1.0385366693192029,
-    "producer_headroom": 13.894240678534976,
+    "producer_batch_seconds": 0.4604758902422077,
+    "producer_p95_seconds": 0.18151160199977312,
+    "gpu_update_seconds": 0.9769783236945297,
+    "producer_batches_per_second": 2.1716663590660645,
+    "gpu_batches_per_second": 1.0235641628346592,
+    "producer_headroom": 2.1216709591041667,
     "sustainable": true,
-    "buffer_seconds": 60.66227785803529,
+    "buffer_seconds": 60.57265606906084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384306.2724841303,
+    "generation_bytes_per_second": 318382.7302578343,
     "upload_bytes_per_second": 4886517.818467076,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.46317988293582,
+    "estimated_full_shards_per_hour": 68.31752234269402,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -76,3 +76,4 @@
 {"step": 727600, "loss": 0.561463475227356, "lr": 5e-05, "grad_norm": 0.21850676530010643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106500106885, "tokens_per_second": 152043.26874381435, "data_wait_s": 1.2847999641962815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727700, "loss": 0.5557624101638794, "lr": 5e-05, "grad_norm": 0.23337268929924065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106514784425, "tokens_per_second": 151908.00363559255, "data_wait_s": 1.4756999917153735e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727800, "loss": 0.5481175184249878, "lr": 5e-05, "grad_norm": 0.23142053883470848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106529601973, "tokens_per_second": 138000.22715871155, "data_wait_s": 2.3218999558594078e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 727900, "loss": 0.5419673919677734, "lr": 5e-05, "grad_norm": 0.22210347484844103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106544291727, "tokens_per_second": 152574.9030902761, "data_wait_s": 1.0648998795659281e-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 sha256:15db3eb4c450c4dcdf0c373e42dcabe6fd85f2db1ba8298a879bb34c619c12d6
-size 166152
+oid sha256:4c343a698c17f1c2bcbbf72cabdca3a55d352d373ef6f628cebae052c7e41d94
+size 168270