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": 77479.13406114948,
-      "required_pages_per_second_upper_estimate": 0.005360186270709692
+      "required_token_positions_per_second": 77402.7761039638,
+      "required_pages_per_second_upper_estimate": 0.005354903650056715
     },
     {
       "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": 9684.891757643685,
-      "required_pages_per_second_upper_estimate": 0.004054861935387632
+      "required_token_positions_per_second": 9675.347012995475,
+      "required_pages_per_second_upper_estimate": 0.00405086575012036
     },
     {
       "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": 19369.78351528737,
-      "required_pages_per_second_upper_estimate": 0.004034215090692204
+      "required_token_positions_per_second": 19350.69402599095,
+      "required_pages_per_second_upper_estimate": 0.004030239253495446
     },
     {
       "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": 29054.675272931057,
-      "required_pages_per_second_upper_estimate": 0.0017417834186918295
+      "required_token_positions_per_second": 29026.041038986423,
+      "required_pages_per_second_upper_estimate": 0.0017400668400887923
     },
     {
       "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": 29054.675272931057,
-      "required_pages_per_second_upper_estimate": 0.0022588732044599743
+      "required_token_positions_per_second": 29026.041038986423,
+      "required_pages_per_second_upper_estimate": 0.002256647018719463
     },
     {
       "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": 29054.675272931057,
-      "required_pages_per_second_upper_estimate": 0.3726821779214102
+      "required_token_positions_per_second": 29026.041038986423,
+      "required_pages_per_second_upper_estimate": 0.3723148887134134
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06361471745316294,
-    "producer_p95_seconds": 0.2269410049993894,
-    "gpu_update_seconds": 0.9515852350880826,
-    "producer_batches_per_second": 15.719632815099137,
-    "gpu_batches_per_second": 1.0508780118971013,
-    "producer_headroom": 14.95857048785445,
+    "producer_batch_seconds": 0.08651116098442913,
+    "producer_p95_seconds": 0.24094921500000055,
+    "gpu_update_seconds": 0.9525239753800561,
+    "producer_batches_per_second": 11.559202172538026,
+    "gpu_batches_per_second": 1.0498423408198214,
+    "producer_headroom": 11.0104172056077,
     "sustainable": true,
-    "buffer_seconds": 60.901455045637285,
+    "buffer_seconds": 60.009010448943535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337896.2153922947,
-    "upload_bytes_per_second": 1915916.5043630984,
+    "generation_bytes_per_second": 357946.5645164134,
+    "upload_bytes_per_second": 1903783.0719585211,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.50466200186376,
+    "estimated_full_shards_per_hour": 76.80700017566014,
     "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": 1.4157619699999486,
-      "bytes": 4898674,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.164695950999885,
       "bytes": 11623258,
@@ -562,6 +556,12 @@
       "bytes": 7396085,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.9337214369988942,
+      "bytes": 1171128,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -197,3 +197,4 @@
 {"step": 664700, "loss": 0.5454566478729248, "lr": 0.0001, "grad_norm": 0.19392334602321665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97295201154, "tokens_per_second": 152737.8006900432, "data_wait_s": 9.757997759152204e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 664800, "loss": 0.5585068464279175, "lr": 0.0001, "grad_norm": 0.18002653357139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97309974586, "tokens_per_second": 153967.55882900333, "data_wait_s": 1.1936997907469049e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 664900, "loss": 0.5658284425735474, "lr": 0.0001, "grad_norm": 0.21101410472547677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97324766629, "tokens_per_second": 154192.23066574917, "data_wait_s": 1.0297000699210912e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 665000, "loss": 0.524940013885498, "lr": 0.0001, "grad_norm": 0.19564966487165172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97339358803, "tokens_per_second": 112068.75674238619, "data_wait_s": 7.923001248855144e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b384c1d6a6e5e4147e795e7f8a0e5d98de98445c86f38ce7e8c74fd0a2f21dd4
-size 423996
+oid sha256:d3245073c83f3d5bcc468eb6ccb9e2ba66f92c7752904d1757e9b2858801500a
+size 426114