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": 76027.63896880443,
-      "required_pages_per_second_upper_estimate": 0.005259768472288636
+      "required_token_positions_per_second": 77568.59306300503,
+      "required_pages_per_second_upper_estimate": 0.005366375252031541
     },
     {
       "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": 9503.454871100554,
-      "required_pages_per_second_upper_estimate": 0.003978898099825057
+      "required_token_positions_per_second": 9696.074132875628,
+      "required_pages_per_second_upper_estimate": 0.004059543762382698
     },
     {
       "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": 19006.909742201107,
-      "required_pages_per_second_upper_estimate": 0.003958638053383261
+      "required_token_positions_per_second": 19392.148265751257,
+      "required_pages_per_second_upper_estimate": 0.004038873078415699
     },
     {
       "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": 28510.36461330166,
-      "required_pages_per_second_upper_estimate": 0.0017091528257613012
+      "required_token_positions_per_second": 29088.222398626887,
+      "required_pages_per_second_upper_estimate": 0.0017437945176537995
     },
     {
       "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": 28510.36461330166,
-      "required_pages_per_second_upper_estimate": 0.0022165554448433564
+      "required_token_positions_per_second": 29088.222398626887,
+      "required_pages_per_second_upper_estimate": 0.0022614813459246133
     },
     {
       "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": 28510.36461330166,
-      "required_pages_per_second_upper_estimate": 0.36570034521493644
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29088.222398626887,
+      "required_pages_per_second_upper_estimate": 0.37311248442973904
     }
   ],
   "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": 0.11819172161744973,
+    "producer_batch_seconds": 0.20161334836757305,
     "producer_p95_seconds": 0.3404495909999241,
-    "gpu_update_seconds": 0.9697525926098006,
-    "producer_batches_per_second": 8.460829458400585,
-    "gpu_batches_per_second": 1.0311908497287927,
-    "producer_headroom": 8.204911302913343,
+    "gpu_update_seconds": 0.9504877823440021,
+    "producer_batches_per_second": 4.959989048824494,
+    "gpu_batches_per_second": 1.0520913772651508,
+    "producer_headroom": 4.71440899146773,
     "sustainable": true,
-    "buffer_seconds": 60.12466074180764,
+    "buffer_seconds": 60.831218070016135,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 466817.32958929247,
-    "upload_bytes_per_second": 5501910.391762454,
+    "generation_bytes_per_second": 545751.6506685159,
+    "upload_bytes_per_second": 5552791.203874074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.16813197859841,
+    "estimated_full_shards_per_hour": 117.10559978524789,
     "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": 1.0192723729996942,
-      "bytes": 2390933,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.0420342719990003,
       "bytes": 6189478,
@@ -562,6 +556,12 @@
       "bytes": 1003302,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 0.9674175760010257,
+      "bytes": 6098431,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -344,3 +344,4 @@
 {"step": 679400, "loss": 0.6518879532814026, "lr": 0.0001, "grad_norm": 0.2733330479010058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99446621460, "tokens_per_second": 153069.31936941762, "data_wait_s": 1.0310999641660601e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679500, "loss": 0.5391449332237244, "lr": 0.0001, "grad_norm": 0.19859551373639187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99461344676, "tokens_per_second": 152318.18885355955, "data_wait_s": 1.2577998859342188e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679600, "loss": 0.5609444975852966, "lr": 0.0001, "grad_norm": 0.18971551553261026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99476059823, "tokens_per_second": 153989.99817463756, "data_wait_s": 9.885996405500919e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 679700, "loss": 0.5336342453956604, "lr": 0.0001, "grad_norm": 0.19925302230334643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99490709871, "tokens_per_second": 152641.51544627736, "data_wait_s": 1.5267003618646413e-05, "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 sha256:3a8a37e117b5fde406abbf51e9960b7bc8b977c5e24c199a8672115ae9781a79
-size 737691
+oid sha256:aa8749118be0d8cc80f9a4867a6a2c0ba59e072bb35d0cd3abe3abf51397edaa
+size 739809