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": 75722.9040660672,
-      "required_pages_per_second_upper_estimate": 0.005238686204634884
+      "required_token_positions_per_second": 77028.96236880489,
+      "required_pages_per_second_upper_estimate": 0.005329042348491056
     },
     {
       "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": 9465.3630082584,
-      "required_pages_per_second_upper_estimate": 0.003962949832301597
+      "required_token_positions_per_second": 9628.620296100611,
+      "required_pages_per_second_upper_estimate": 0.004031302249521287
     },
     {
       "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": 18930.7260165168,
-      "required_pages_per_second_upper_estimate": 0.003942770992423172
+      "required_token_positions_per_second": 19257.240592201222,
+      "required_pages_per_second_upper_estimate": 0.004010775367769827
     },
     {
       "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": 28396.0890247752,
-      "required_pages_per_second_upper_estimate": 0.0017023021787178603
+      "required_token_positions_per_second": 28885.860888301835,
+      "required_pages_per_second_upper_estimate": 0.0017316632540979478
     },
     {
       "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": 28396.0890247752,
-      "required_pages_per_second_upper_estimate": 0.0022076710204806172
+      "required_token_positions_per_second": 28885.860888301835,
+      "required_pages_per_second_upper_estimate": 0.0022457486285910565
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28396.0890247752,
-      "required_pages_per_second_upper_estimate": 0.3642345406648863
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28885.860888301835,
+      "required_pages_per_second_upper_estimate": 0.3705168082541506
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1257561896092625,
-    "producer_p95_seconds": 0.3574241579990485,
-    "gpu_update_seconds": 0.9736552092042499,
-    "producer_batches_per_second": 7.951894877755946,
-    "gpu_batches_per_second": 1.0270576180835937,
-    "producer_headroom": 7.742403870771669,
+    "producer_batch_seconds": 0.09918057738303787,
+    "producer_p95_seconds": 0.3622318310008268,
+    "gpu_update_seconds": 0.957146477541781,
+    "producer_batches_per_second": 10.082619262620089,
+    "gpu_batches_per_second": 1.0447721675456392,
+    "producer_headroom": 9.650543511611728,
     "sustainable": true,
-    "buffer_seconds": 60.36662297066349,
+    "buffer_seconds": 60.3002280851322,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 523263.4243114643,
+    "generation_bytes_per_second": 518088.4853541351,
     "upload_bytes_per_second": 4519218.487224701,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.28014990814158,
+    "estimated_full_shards_per_hour": 111.16972847431221,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -433,3 +433,4 @@
 {"step": 803300, "loss": 0.5157017707824707, "lr": 3e-05, "grad_norm": 0.2845687018671663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117579791596, "tokens_per_second": 152538.1604435676, "data_wait_s": 1.0247997124679387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803400, "loss": 0.5423639416694641, "lr": 3e-05, "grad_norm": 0.2784970674390396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117594576659, "tokens_per_second": 153697.22625140508, "data_wait_s": 8.11300560599193e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 803500, "loss": 0.5135219693183899, "lr": 3e-05, "grad_norm": 0.28931642332439106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117609406982, "tokens_per_second": 154380.1701985896, "data_wait_s": 2.8823000320699066e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 803600, "loss": 0.5232123732566833, "lr": 3e-05, "grad_norm": 0.2649226164007285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117623837178, "tokens_per_second": 147991.02297028588, "data_wait_s": 1.6224999853875488e-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 sha256:7f198379956ffd04885fb23c56d7cfbf659449832ad6aebcb3ba5d1526e166cc
-size 927759
+oid sha256:944d7d6c378dc5e25e32898f752c1f0890ab6ebeed47a1a0cc38f90a4d73d4da
+size 929877