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": 75694.45947528123,
-      "required_pages_per_second_upper_estimate": 0.005236718341843766
+      "required_token_positions_per_second": 77423.96757010218,
+      "required_pages_per_second_upper_estimate": 0.005356369724855144
     },
     {
       "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": 9461.807434410153,
-      "required_pages_per_second_upper_estimate": 0.003961461187780161
+      "required_token_positions_per_second": 9677.995946262772,
+      "required_pages_per_second_upper_estimate": 0.00405197480316336
     },
     {
       "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": 18923.614868820307,
-      "required_pages_per_second_upper_estimate": 0.003941289927891568
+      "required_token_positions_per_second": 19355.991892525544,
+      "required_pages_per_second_upper_estimate": 0.004031342659380466
     },
     {
       "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": 28385.42230323046,
-      "required_pages_per_second_upper_estimate": 0.0017016627250483942
+      "required_token_positions_per_second": 29033.98783878832,
+      "required_pages_per_second_upper_estimate": 0.0017405432386028551
     },
     {
       "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": 28385.42230323046,
-      "required_pages_per_second_upper_estimate": 0.002206841729798463
+      "required_token_positions_per_second": 29033.98783878832,
+      "required_pages_per_second_upper_estimate": 0.0022572648474497824
     },
     {
       "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": 28385.42230323046,
-      "required_pages_per_second_upper_estimate": 0.36409771941394364
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29033.98783878832,
+      "required_pages_per_second_upper_estimate": 0.3724168217286633
     }
   ],
   "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": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3224483322966307,
-    "producer_p95_seconds": 0.37007798500417266,
-    "gpu_update_seconds": 0.9740210909898446,
-    "producer_batches_per_second": 3.101272048385313,
-    "gpu_batches_per_second": 1.026671813629574,
-    "producer_headroom": 3.0207043840245724,
+    "producer_batch_seconds": 0.350425369203208,
+    "producer_p95_seconds": 0.3773666150009376,
+    "gpu_update_seconds": 0.9522632630941351,
+    "producer_batches_per_second": 2.8536746705119698,
+    "gpu_batches_per_second": 1.0501297684746933,
+    "producer_headroom": 2.7174495535508094,
     "sustainable": true,
-    "buffer_seconds": 60.389307641370365,
+    "buffer_seconds": 60.94484883802465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 479393.3940155107,
-    "upload_bytes_per_second": 5045484.191235549,
+    "generation_bytes_per_second": 494837.46299891575,
+    "upload_bytes_per_second": 5127793.568167325,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.8666626486682,
+    "estimated_full_shards_per_hour": 106.18060033298116,
     "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": 1.342605496996839,
-      "bytes": 1094629,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.425732910996885,
       "bytes": 11302182,
@@ -562,6 +556,12 @@
       "bytes": 10468749,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.4208347750027315,
+      "bytes": 8470292,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -439,3 +439,4 @@
 {"step": 968900, "loss": 0.5428612232208252, "lr": 3e-05, "grad_norm": 0.3556979921311017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141820691943, "tokens_per_second": 151478.29184310292, "data_wait_s": 1.683599839452654e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 969000, "loss": 0.4934856593608856, "lr": 3e-05, "grad_norm": 0.3608491779152761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141835475142, "tokens_per_second": 153380.01690895387, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 969100, "loss": 0.5850196480751038, "lr": 3e-05, "grad_norm": 0.3727238963613872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141850153217, "tokens_per_second": 152703.9038187476, "data_wait_s": 9.55599534790963e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 969200, "loss": 0.5263361930847168, "lr": 3e-05, "grad_norm": 0.33747269269095503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141864800105, "tokens_per_second": 152094.22411823884, "data_wait_s": 8.386996341869235e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53ab84d17364472fc34f8c49b406e805c98f654260d87f4e10a8c5c911ac762d
-size 940467
+oid sha256:82e3efa762ab45007a191cc8d3bab8becd918ee08d28d9d66bfa27f30ccf7ce2
+size 942585