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": 77246.60067499607,
-      "required_pages_per_second_upper_estimate": 0.005344099071503807
+      "required_token_positions_per_second": 76862.36435195869,
+      "required_pages_per_second_upper_estimate": 0.00531751671112498
     },
     {
       "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": 9655.825084374508,
-      "required_pages_per_second_upper_estimate": 0.004042692326270987
+      "required_token_positions_per_second": 9607.795543994836,
+      "required_pages_per_second_upper_estimate": 0.004022583360684874
     },
     {
       "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": 19311.650168749016,
-      "required_pages_per_second_upper_estimate": 0.004022107447687708
+      "required_token_positions_per_second": 19215.591087989673,
+      "required_pages_per_second_upper_estimate": 0.00400210087441385
     },
     {
       "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": 28967.475253123528,
-      "required_pages_per_second_upper_estimate": 0.001736555910651089
+      "required_token_positions_per_second": 28823.38663198451,
+      "required_pages_per_second_upper_estimate": 0.0017279180178243903
     },
     {
       "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": 28967.475253123528,
-      "required_pages_per_second_upper_estimate": 0.0022520937864722907
+      "required_token_positions_per_second": 28823.38663198451,
+      "required_pages_per_second_upper_estimate": 0.002240891529957597
     },
     {
       "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": 28967.475253123528,
-      "required_pages_per_second_upper_estimate": 0.3715636696954058
+      "required_token_positions_per_second": 28823.38663198451,
+      "required_pages_per_second_upper_estimate": 0.36971545557374214
     }
   ],
   "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": 0.17778230753881985,
-    "producer_p95_seconds": 0.2639045929972781,
-    "gpu_update_seconds": 0.9544497667955633,
-    "producer_batches_per_second": 5.624856679181329,
-    "gpu_batches_per_second": 1.047724075995498,
-    "producer_headroom": 5.368643145703086,
+    "producer_batch_seconds": 0.09111212314832073,
+    "producer_p95_seconds": 0.2562402639996435,
+    "gpu_update_seconds": 0.9592210781129995,
+    "producer_batches_per_second": 10.975487843391681,
+    "gpu_batches_per_second": 1.0425125373258286,
+    "producer_headroom": 10.527919281954288,
     "sustainable": true,
-    "buffer_seconds": 60.130335308120486,
+    "buffer_seconds": 60.43092792111897,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424466.44954264065,
+    "generation_bytes_per_second": 524429.3008187664,
     "upload_bytes_per_second": 4983425.0524334675,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.08061899861731,
+    "estimated_full_shards_per_hour": 112.53031986639255,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 905400, "loss": 0.542047381401062, "lr": 3e-05, "grad_norm": 0.33252034845310646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132525219983, "tokens_per_second": 153165.92505290068, "data_wait_s": 1.596699803485535e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 905500, "loss": 0.5168270468711853, "lr": 3e-05, "grad_norm": 0.32103477269765957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132539846636, "tokens_per_second": 152342.34680555415, "data_wait_s": 1.6169000446097925e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 905600, "loss": 0.5333682298660278, "lr": 3e-05, "grad_norm": 0.34868177279533796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132554316032, "tokens_per_second": 150367.3079037568, "data_wait_s": 1.9656003132695332e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 905700, "loss": 0.5601820945739746, "lr": 3e-05, "grad_norm": 0.318668653213803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132568963580, "tokens_per_second": 152217.31002958657, "data_wait_s": 1.2447999324649572e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b98bfd4e1ccdb77672e52f9779e9e2b41e7ff0e4405138aaf8d78601e4e9f1c
-size 439605
+oid sha256:a3b1f1c6f20bdd2a2948b179b8ad46d99e5813583ebc069a106cd0be6989dd73
+size 441723