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": 76252.76973508314,
-      "required_pages_per_second_upper_estimate": 0.005275343541075145
+      "required_token_positions_per_second": 77239.71489432867,
+      "required_pages_per_second_upper_estimate": 0.005343622697220024
     },
     {
       "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": 9531.596216885393,
-      "required_pages_per_second_upper_estimate": 0.003990680293647044
+      "required_token_positions_per_second": 9654.964361791084,
+      "required_pages_per_second_upper_estimate": 0.004042331959699239
     },
     {
       "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": 19063.192433770786,
-      "required_pages_per_second_upper_estimate": 0.003970360253762835
+      "required_token_positions_per_second": 19309.92872358217,
+      "required_pages_per_second_upper_estimate": 0.004021748916057015
     },
     {
       "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": 28594.78865065618,
-      "required_pages_per_second_upper_estimate": 0.0017142139178926638
+      "required_token_positions_per_second": 28964.893085373253,
+      "required_pages_per_second_upper_estimate": 0.0017364011136372017
     },
     {
       "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": 28594.78865065618,
-      "required_pages_per_second_upper_estimate": 0.002223119042405575
+      "required_token_positions_per_second": 28964.893085373253,
+      "required_pages_per_second_upper_estimate": 0.0022518930342874627
     },
     {
       "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": 28594.78865065618,
-      "required_pages_per_second_upper_estimate": 0.3667832461186514
+      "required_token_positions_per_second": 28964.893085373253,
+      "required_pages_per_second_upper_estimate": 0.3715305484200209
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12214488600025675,
-    "producer_p95_seconds": 0.292323795001721,
-    "gpu_update_seconds": 0.9668894684894112,
-    "producer_batches_per_second": 8.186998512552528,
-    "gpu_batches_per_second": 1.0342443811724602,
-    "producer_headroom": 7.915922640325513,
+    "producer_batch_seconds": 0.4318656000080523,
+    "producer_p95_seconds": 0.2652624959991954,
+    "gpu_update_seconds": 0.9545348542633406,
+    "producer_batches_per_second": 2.3155352035016326,
+    "gpu_batches_per_second": 1.0476306816179561,
+    "producer_headroom": 2.2102590580160655,
     "sustainable": true,
-    "buffer_seconds": 60.9140365148329,
+    "buffer_seconds": 60.135695818590456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377696.323247973,
-    "upload_bytes_per_second": 5840976.5976948235,
+    "generation_bytes_per_second": 324639.36622525955,
+    "upload_bytes_per_second": 5746466.857091533,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.04483864859955,
+    "estimated_full_shards_per_hour": 69.66005077427235,
     "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": 1.1790409999994154,
-      "bytes": 12018438,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0833680409996305,
       "bytes": 2366981,
@@ -562,6 +556,12 @@
       "bytes": 2592499,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.24596934100191,
+      "bytes": 5407802,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 583500, "loss": 0.618613064289093, "lr": 0.0003, "grad_norm": 0.1872197076454018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85407741399, "tokens_per_second": 153216.23394063968, "data_wait_s": 1.3447999663185328e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 583600, "loss": 0.5512357354164124, "lr": 0.0003, "grad_norm": 0.22134689228099375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85422320440, "tokens_per_second": 150647.1422160129, "data_wait_s": 1.1094998626504093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 583700, "loss": 0.5807870030403137, "lr": 0.0003, "grad_norm": 0.14827346223682342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85436819915, "tokens_per_second": 150701.56385398054, "data_wait_s": 6.705999112455174e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 583800, "loss": 0.5811905860900879, "lr": 0.0003, "grad_norm": 0.14513462240666394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85451400080, "tokens_per_second": 133331.34935116593, "data_wait_s": 1.553099718876183e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:31b156ac6d76a10ef1c384033fd3eeb472b63546c031b8097eb5c8b1747f5a5b
-size 567468
+oid sha256:b5bc0ed41e21e710c86cf158fc217da87b35e00353598941c8a0450d383ff5cf
+size 569586