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": 76160.4693449393,
-      "required_pages_per_second_upper_estimate": 0.0052689579859185815
+      "required_token_positions_per_second": 75980.05818543474,
+      "required_pages_per_second_upper_estimate": 0.005256476723292493
     },
     {
       "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": 9520.058668117412,
-      "required_pages_per_second_upper_estimate": 0.003985849762909306
+      "required_token_positions_per_second": 9497.507273179343,
+      "required_pages_per_second_upper_estimate": 0.003976407964775414
     },
     {
       "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": 19040.117336234824,
-      "required_pages_per_second_upper_estimate": 0.003965554319477338
+      "required_token_positions_per_second": 18995.014546358685,
+      "required_pages_per_second_upper_estimate": 0.003956160597786694
     },
     {
       "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": 28560.176004352237,
-      "required_pages_per_second_upper_estimate": 0.001712138942597193
+      "required_token_positions_per_second": 28492.521819538026,
+      "required_pages_per_second_upper_estimate": 0.0017080831775195399
     },
     {
       "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": 28560.176004352237,
-      "required_pages_per_second_upper_estimate": 0.0022204280614003848
+      "required_token_positions_per_second": 28492.521819538026,
+      "required_pages_per_second_upper_estimate": 0.0022151682461103894
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28560.176004352237,
-      "required_pages_per_second_upper_estimate": 0.36633927225602847
+      "required_token_positions_per_second": 28492.521819538026,
+      "required_pages_per_second_upper_estimate": 0.36547147701463584
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3790672087188227,
-    "producer_p95_seconds": 0.3114004990020476,
-    "gpu_update_seconds": 0.9680612610996083,
-    "producer_batches_per_second": 2.6380546166992804,
-    "gpu_batches_per_second": 1.0329924770092678,
-    "producer_headroom": 2.553798479091549,
+    "producer_batch_seconds": 0.10464862774210815,
+    "producer_p95_seconds": 0.3031370139979117,
+    "gpu_update_seconds": 0.970359878115144,
+    "producer_batches_per_second": 9.555787033006869,
+    "gpu_batches_per_second": 1.030545494051578,
+    "producer_headroom": 9.27255234064282,
     "sustainable": true,
-    "buffer_seconds": 60.01979818817571,
+    "buffer_seconds": 60.16231244313893,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338454.5184263395,
-    "upload_bytes_per_second": 5383029.600886166,
+    "generation_bytes_per_second": 357923.04263373284,
+    "upload_bytes_per_second": 5423002.84076213,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.62446083634032,
+    "estimated_full_shards_per_hour": 76.8019529271983,
     "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": 1.0160050430022238,
-      "bytes": 1159604,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.8482585339988873,
       "bytes": 12054418,
@@ -562,6 +556,12 @@
       "bytes": 7404435,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.136316104002617,
+      "bytes": 5061136,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 469200, "loss": 0.5728943347930908, "lr": 0.0003, "grad_norm": 0.14785872656051363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68676850819, "tokens_per_second": 150585.90830807685, "data_wait_s": 1.7526999727124348e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 469300, "loss": 0.5722887516021729, "lr": 0.0003, "grad_norm": 0.15890479592170773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68691498974, "tokens_per_second": 151630.33289074, "data_wait_s": 1.0546002158662304e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 469400, "loss": 0.5479133129119873, "lr": 0.0003, "grad_norm": 0.1530461936897756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68706157174, "tokens_per_second": 152615.7330602104, "data_wait_s": 1.4416000340133905e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 469500, "loss": 0.5701079964637756, "lr": 0.0003, "grad_norm": 0.14553400770053482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68720786596, "tokens_per_second": 151893.27550886673, "data_wait_s": 1.0511997970752418e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9712bdd63757ba8b69137250457b5164d65aaf41792f282993a628813580f43
-size 622536
+oid sha256:f9bfc38ad6fe81d1266fc7ad787c285072a31966fd5e595ca0f9ee950b1b86b4
+size 624654