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": 75818.36498344786,
-      "required_pages_per_second_upper_estimate": 0.005245290412399122
+      "required_token_positions_per_second": 77240.65851921892,
+      "required_pages_per_second_upper_estimate": 0.005343687979379443
     },
     {
       "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": 9477.295622930982,
-      "required_pages_per_second_upper_estimate": 0.003967945768883677
+      "required_token_positions_per_second": 9655.082314902365,
+      "required_pages_per_second_upper_estimate": 0.004042381344203792
     },
     {
       "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": 18954.591245861964,
-      "required_pages_per_second_upper_estimate": 0.0039477414903273484
+      "required_token_positions_per_second": 19310.16462980473,
+      "required_pages_per_second_upper_estimate": 0.00402179804910191
     },
     {
       "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": 28431.886868792946,
-      "required_pages_per_second_upper_estimate": 0.0017044482048065813
+      "required_token_positions_per_second": 28965.246944707098,
+      "required_pages_per_second_upper_estimate": 0.0017364223269639538
     },
     {
       "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": 28431.886868792946,
-      "required_pages_per_second_upper_estimate": 0.0022104541453949215
+      "required_token_positions_per_second": 28965.246944707098,
+      "required_pages_per_second_upper_estimate": 0.002251920545294206
     },
     {
       "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": 28431.886868792946,
-      "required_pages_per_second_upper_estimate": 0.36469371697121566
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28965.246944707098,
+      "required_pages_per_second_upper_estimate": 0.371535087347611
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08974102178086696,
-    "producer_p95_seconds": 0.3063213339992217,
-    "gpu_update_seconds": 0.9724293054340566,
-    "producer_batches_per_second": 11.143175998617869,
-    "gpu_batches_per_second": 1.0283523896409485,
-    "producer_headroom": 10.835950896665423,
+    "producer_batch_seconds": 0.37331648620335045,
+    "producer_p95_seconds": 0.3498321989973192,
+    "gpu_update_seconds": 0.954523193010519,
+    "producer_batches_per_second": 2.6786923078861475,
+    "gpu_batches_per_second": 1.047643480349649,
+    "producer_headroom": 2.5568739348162017,
     "sustainable": true,
-    "buffer_seconds": 60.29061693691151,
+    "buffer_seconds": 60.1349611596627,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405225.38388075685,
-    "upload_bytes_per_second": 4944662.185689701,
+    "generation_bytes_per_second": 374147.0167325487,
+    "upload_bytes_per_second": 4950259.128539045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.95193421666173,
+    "estimated_full_shards_per_hour": 80.28324009401652,
     "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": 0.9436758240008203,
-      "bytes": 855631,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.5082043110014638,
       "bytes": 16602529,
@@ -562,6 +556,12 @@
       "bytes": 7561561,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9214401089993771,
+      "bytes": 1231310,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 920400, "loss": 0.5290529131889343, "lr": 3e-05, "grad_norm": 0.322497223396615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134719629995, "tokens_per_second": 152717.83267476334, "data_wait_s": 1.0044001101050526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 920500, "loss": 0.49328935146331787, "lr": 3e-05, "grad_norm": 0.324737142035647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134734212845, "tokens_per_second": 152135.00085540628, "data_wait_s": 1.0896001185756177e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 920600, "loss": 0.47619855403900146, "lr": 3e-05, "grad_norm": 0.32074856379528593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134748899382, "tokens_per_second": 153541.57483697246, "data_wait_s": 1.5821999113541096e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 920700, "loss": 0.5966166257858276, "lr": 3e-05, "grad_norm": 0.4249011900687419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134763610364, "tokens_per_second": 153825.9132291072, "data_wait_s": 1.3163000403437763e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0977d9d2d9fc32d26188c594e701bf9eda7218c5c789712f349d7647f77e9a0a
-size 759654
+oid sha256:e7556afa62271b0faadf1754984f66c24635d170c25000c0f598abb52654ce8e
+size 761772