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": 76775.09924471528,
-      "required_pages_per_second_upper_estimate": 0.005311479508522936
+      "required_token_positions_per_second": 77253.63815372821,
+      "required_pages_per_second_upper_estimate": 0.005344585940611697
     },
     {
       "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": 9596.88740558941,
-      "required_pages_per_second_upper_estimate": 0.004018016350922355
+      "required_token_positions_per_second": 9656.704769216027,
+      "required_pages_per_second_upper_estimate": 0.004043060631944223
     },
     {
       "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": 19193.77481117882,
-      "required_pages_per_second_upper_estimate": 0.003997557119288058
+      "required_token_positions_per_second": 19313.409538432054,
+      "required_pages_per_second_upper_estimate": 0.004022473877994986
     },
     {
       "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": 28790.662216768233,
-      "required_pages_per_second_upper_estimate": 0.0017259562390994632
+      "required_token_positions_per_second": 28970.114307648077,
+      "required_pages_per_second_upper_estimate": 0.0017367141179402325
     },
     {
       "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": 28790.662216768233,
-      "required_pages_per_second_upper_estimate": 0.0022383473506140217
+      "required_token_positions_per_second": 28970.114307648077,
+      "required_pages_per_second_upper_estimate": 0.0022522989613536003
     },
     {
       "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": 28790.662216768233,
-      "required_pages_per_second_upper_estimate": 0.3692957019120872
+      "required_token_positions_per_second": 28970.114307648077,
+      "required_pages_per_second_upper_estimate": 0.3715975206532507
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.41159245849216575,
-    "producer_p95_seconds": 0.17366555700027675,
-    "gpu_update_seconds": 0.9603113603929984,
-    "producer_batches_per_second": 2.4295877617957715,
-    "gpu_batches_per_second": 1.0413289285578788,
-    "producer_headroom": 2.3331607287242773,
+    "producer_batch_seconds": 0.06781120253121742,
+    "producer_p95_seconds": 0.18212803899950814,
+    "gpu_update_seconds": 0.9543628204705067,
+    "producer_batches_per_second": 14.746825932479846,
+    "gpu_batches_per_second": 1.047819527909725,
+    "producer_headroom": 14.073822389909076,
     "sustainable": true,
-    "buffer_seconds": 60.499615704758895,
+    "buffer_seconds": 60.12485768964192,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278039.48185573175,
-    "upload_bytes_per_second": 4417463.248551184,
+    "generation_bytes_per_second": 277842.28489207913,
+    "upload_bytes_per_second": 4432180.270266359,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.660800378360406,
+    "estimated_full_shards_per_hour": 59.61848650047093,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 13477323,
       "pages": 45,
       "shards": 4
+    },
+    {
+      "seconds": 1.0930595740010176,
+      "bytes": 6005049,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -80,3 +80,4 @@
 {"step": 533000, "loss": 0.565763533115387, "lr": 0.0003, "grad_norm": 0.14975142137259417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78014125467, "tokens_per_second": 152246.60469046727, "data_wait_s": 1.406900082656648e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 533100, "loss": 0.5684452652931213, "lr": 0.0003, "grad_norm": 0.14275307081191532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78028819236, "tokens_per_second": 145565.77981952298, "data_wait_s": 1.0923999070655555e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 533200, "loss": 0.5814388394355774, "lr": 0.0003, "grad_norm": 0.15108192271901014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78043485668, "tokens_per_second": 153120.8393118724, "data_wait_s": 2.0298000890761614e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 533300, "loss": 0.546099066734314, "lr": 0.0003, "grad_norm": 0.16129710233637432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78058151537, "tokens_per_second": 153524.2972613156, "data_wait_s": 8.629000149085186e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcae0a217d0244fde9e63e4309f90b13c8abfb992b636df9758189d9eadb09ae
-size 174624
+oid sha256:5bc35ee7aec6230710d733601626b9279f83fb4a5d4deb2928aeda2978ed52d6
+size 176742