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": 77365.33532253442,
-      "required_pages_per_second_upper_estimate": 0.005352313409922849
+      "required_token_positions_per_second": 75994.03182103377,
+      "required_pages_per_second_upper_estimate": 0.005257443451826536
     },
     {
       "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": 9670.666915316802,
-      "required_pages_per_second_upper_estimate": 0.004048906290953853
+      "required_token_positions_per_second": 9499.253977629221,
+      "required_pages_per_second_upper_estimate": 0.003977139273453241
     },
     {
       "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": 19341.333830633605,
-      "required_pages_per_second_upper_estimate": 0.004028289771647462
+      "required_token_positions_per_second": 18998.507955258443,
+      "required_pages_per_second_upper_estimate": 0.003956888182733125
     },
     {
       "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": 29012.00074595041,
-      "required_pages_per_second_upper_estimate": 0.0017392251459595694
+      "required_token_positions_per_second": 28497.761932887664,
+      "required_pages_per_second_upper_estimate": 0.001708397314313661
     },
     {
       "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": 29012.00074595041,
-      "required_pages_per_second_upper_estimate": 0.002255555447692636
+      "required_token_positions_per_second": 28497.761932887664,
+      "required_pages_per_second_upper_estimate": 0.002215575641874501
     },
     {
       "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": 29012.00074595041,
-      "required_pages_per_second_upper_estimate": 0.3721347949096395
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28497.761932887664,
+      "required_pages_per_second_upper_estimate": 0.3655386915622897
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "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": 0.10793024863289702,
-    "producer_p95_seconds": 0.2442270199990162,
-    "gpu_update_seconds": 0.9529849472328861,
-    "producer_batches_per_second": 9.265243179428767,
-    "gpu_batches_per_second": 1.0493345177210072,
-    "producer_headroom": 8.82963728244778,
+    "producer_batch_seconds": 0.6874219265625072,
+    "producer_p95_seconds": 0.24215357400134963,
+    "gpu_update_seconds": 0.9701814502174291,
+    "producer_batches_per_second": 1.4547106534709437,
+    "gpu_batches_per_second": 1.0307350236142818,
+    "producer_headroom": 1.4113332914311842,
     "sustainable": true,
-    "buffer_seconds": 60.038051675671824,
+    "buffer_seconds": 60.151249913480605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398086.69224948075,
-    "upload_bytes_per_second": 5232468.799184691,
+    "generation_bytes_per_second": 319243.8606824047,
+    "upload_bytes_per_second": 5286994.596617914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.42013717282597,
+    "estimated_full_shards_per_hour": 68.5023008857165,
     "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": [
-    {
-      "seconds": 1.1932894399997167,
-      "bytes": 6531618,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1864704220006388,
-      "bytes": 6136727,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2697396670000671,
       "bytes": 6182466,
@@ -562,6 +550,18 @@
       "bytes": 891680,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.5061208509996504,
+      "bytes": 11821724,
+      "pages": 33,
+      "shards": 4
+    },
+    {
+      "seconds": 1.1246318519988563,
+      "bytes": 6660673,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 899800, "loss": 0.588070809841156, "lr": 3e-05, "grad_norm": 0.340170300486688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131705363330, "tokens_per_second": 151823.2719513825, "data_wait_s": 1.5900999642326497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899900, "loss": 0.5504563450813293, "lr": 3e-05, "grad_norm": 0.32135612538305525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131719837800, "tokens_per_second": 151262.79724883498, "data_wait_s": 1.4798999472986907e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 900000, "loss": 0.5382843017578125, "lr": 3e-05, "grad_norm": 0.327387548139761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131734420906, "tokens_per_second": 134785.8818095809, "data_wait_s": 8.85099871084094e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 900100, "loss": 0.5342219471931458, "lr": 3e-05, "grad_norm": 0.3243542243782647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131749066860, "tokens_per_second": 134505.8787989976, "data_wait_s": 9.325000064563937e-06, "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:eb173dbd4c3b8d2ed6bd3478a9ea895cf7ebf20bb4bf6900a88f67dd340628b6
-size 320214
+oid sha256:4f1a7ec8774030feb5e1658d362766bd52734e5b341966fff0da119779b8937e
+size 322332