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": 77055.04594382628,
-      "required_pages_per_second_upper_estimate": 0.005330846870733269
+      "required_token_positions_per_second": 76993.93903629287,
+      "required_pages_per_second_upper_estimate": 0.005326619353082548
     },
     {
       "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": 9631.880742978285,
-      "required_pages_per_second_upper_estimate": 0.004032667330543096
+      "required_token_positions_per_second": 9624.242379536608,
+      "required_pages_per_second_upper_estimate": 0.004029469307277233
     },
     {
       "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": 19263.76148595657,
-      "required_pages_per_second_upper_estimate": 0.004012133497971533
+      "required_token_positions_per_second": 19248.484759073217,
+      "required_pages_per_second_upper_estimate": 0.004008951758636112
     },
     {
       "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": 28895.642228934852,
-      "required_pages_per_second_upper_estimate": 0.0017322496305336547
+      "required_token_positions_per_second": 28872.727138609825,
+      "required_pages_per_second_upper_estimate": 0.0017308759058579865
     },
     {
       "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": 28895.642228934852,
-      "required_pages_per_second_upper_estimate": 0.002246509084801703
+      "required_token_positions_per_second": 28872.727138609825,
+      "required_pages_per_second_upper_estimate": 0.0022447275373217624
     },
     {
       "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": 28895.642228934852,
-      "required_pages_per_second_upper_estimate": 0.37064227278940565
+      "required_token_positions_per_second": 28872.727138609825,
+      "required_pages_per_second_upper_estimate": 0.3703483426150232
     }
   ],
   "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": 0.12771181899222483,
-    "producer_p95_seconds": 0.27902829999948153,
-    "gpu_update_seconds": 0.9568224779691687,
-    "producer_batches_per_second": 7.830128862708318,
-    "gpu_batches_per_second": 1.0451259486738589,
-    "producer_headroom": 7.492043301234482,
+    "producer_batch_seconds": 0.07570984789836643,
+    "producer_p95_seconds": 0.2726982069980295,
+    "gpu_update_seconds": 0.957581868427937,
+    "producer_batches_per_second": 13.208321344700215,
+    "gpu_batches_per_second": 1.0442971331962465,
+    "producer_headroom": 12.648049032054633,
     "sustainable": true,
-    "buffer_seconds": 60.27981611205763,
+    "buffer_seconds": 60.32765771096003,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 605901.0345789953,
-    "upload_bytes_per_second": 4770624.437406828,
+    "generation_bytes_per_second": 606017.5448857261,
+    "upload_bytes_per_second": 4733091.295151953,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.00612868322084,
+    "estimated_full_shards_per_hour": 65.01862888302249,
     "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": 1.591609420000168,
-      "bytes": 11822162,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9329451240009803,
       "bytes": 2485686,
@@ -562,6 +556,12 @@
       "bytes": 6289450,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1195497589978913,
+      "bytes": 6194218,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -247,3 +247,4 @@
 {"step": 464900, "loss": 0.5823112726211548, "lr": 0.0003, "grad_norm": 0.1557854975686474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68047434277, "tokens_per_second": 153518.83346817302, "data_wait_s": 1.333300315309316e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465000, "loss": 0.6226056218147278, "lr": 0.0003, "grad_norm": 0.1717177493832919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68062086371, "tokens_per_second": 151632.5618039361, "data_wait_s": 1.2479998986236751e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465100, "loss": 0.593877375125885, "lr": 0.0003, "grad_norm": 0.1601375107492322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68076687952, "tokens_per_second": 137693.72199486068, "data_wait_s": 8.839000656735152e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 465200, "loss": 0.5925023555755615, "lr": 0.0003, "grad_norm": 0.16925174414235936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68091189996, "tokens_per_second": 150627.41191968432, "data_wait_s": 1.1931002518394962e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fd7dc72fb32bf1e0088d6b205257d0224342f16df3c29e22f7fb16b76c295a6
-size 531462
+oid sha256:536bf99144d7d8f124bccc6e97ed76854a9d0753cc905e782e1405198ae5f50b
+size 533580