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": 77587.46580255701,
-      "required_pages_per_second_upper_estimate": 0.005367680911944536
+      "required_token_positions_per_second": 76775.09924471528,
+      "required_pages_per_second_upper_estimate": 0.005311479508522936
     },
     {
       "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": 9698.433225319626,
-      "required_pages_per_second_upper_estimate": 0.0040605314651255475
+      "required_token_positions_per_second": 9596.88740558941,
+      "required_pages_per_second_upper_estimate": 0.004018016350922355
     },
     {
       "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": 19396.866450639252,
-      "required_pages_per_second_upper_estimate": 0.004039855751900966
+      "required_token_positions_per_second": 19193.77481117882,
+      "required_pages_per_second_upper_estimate": 0.003997557119288058
     },
     {
       "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": 29095.299675958882,
-      "required_pages_per_second_upper_estimate": 0.0017442187896234758
+      "required_token_positions_per_second": 28790.662216768233,
+      "required_pages_per_second_upper_estimate": 0.0017259562390994632
     },
     {
       "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": 29095.299675958882,
-      "required_pages_per_second_upper_estimate": 0.002262031573107523
+      "required_token_positions_per_second": 28790.662216768233,
+      "required_pages_per_second_upper_estimate": 0.0022383473506140217
     },
     {
       "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": 29095.299675958882,
-      "required_pages_per_second_upper_estimate": 0.37320326414436555
+      "required_token_positions_per_second": 28790.662216768233,
+      "required_pages_per_second_upper_estimate": 0.3692957019120872
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.10000781278903048,
-    "producer_p95_seconds": 0.1768303390017536,
-    "gpu_update_seconds": 0.9502565812320961,
-    "producer_batches_per_second": 9.999218782131855,
-    "gpu_batches_per_second": 1.052347355177911,
-    "producer_headroom": 9.50182345490038,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.81642119885415,
+    "buffer_seconds": 60.499615704758895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281731.86079268827,
-    "upload_bytes_per_second": 4325756.376371411,
+    "generation_bytes_per_second": 278039.48185573175,
+    "upload_bytes_per_second": 4417463.248551184,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.45309894404875,
+    "estimated_full_shards_per_hour": 59.660800378360406,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 757336,
       "pages": 24,
       "shards": 1
+    },
+    {
+      "seconds": 1.4440002779992938,
+      "bytes": 13477323,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"step": 532900, "loss": 0.5626486539840698, "lr": 0.0003, "grad_norm": 0.14230132480504046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77999558782, "tokens_per_second": 152497.5026469427, "data_wait_s": 1.5833000361453742e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:9e4df1cbb566145033cd9c89f73c4a5710d89ff53ab6979a392acf6ac98d0539
-size 172506
+oid sha256:dcae0a217d0244fde9e63e4309f90b13c8abfb992b636df9758189d9eadb09ae
+size 174624