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": 76418.14584616253,
-      "required_pages_per_second_upper_estimate": 0.005286784644165062
+      "required_token_positions_per_second": 75643.89261973655,
+      "required_pages_per_second_upper_estimate": 0.005233220009445916
     },
     {
       "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": 9552.268230770316,
-      "required_pages_per_second_upper_estimate": 0.003999335234179923
+      "required_token_positions_per_second": 9455.486577467069,
+      "required_pages_per_second_upper_estimate": 0.003958814776972594
     },
     {
       "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": 19104.536461540632,
-      "required_pages_per_second_upper_estimate": 0.003978971124431962
+      "required_token_positions_per_second": 18910.973154934138,
+      "required_pages_per_second_upper_estimate": 0.003938656992273494
     },
     {
       "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": 28656.80469231095,
-      "required_pages_per_second_upper_estimate": 0.0017179316848968545
+      "required_token_positions_per_second": 28366.45973240121,
+      "required_pages_per_second_upper_estimate": 0.0017005259478813496
     },
     {
       "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": 28656.80469231095,
-      "required_pages_per_second_upper_estimate": 0.002227940516864494
+      "required_token_positions_per_second": 28366.45973240121,
+      "required_pages_per_second_upper_estimate": 0.0022053674733240234
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28656.80469231095,
-      "required_pages_per_second_upper_estimate": 0.36757872131336117
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28366.45973240121,
+      "required_pages_per_second_upper_estimate": 0.3638544879157683
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12741619256237868,
-    "producer_p95_seconds": 0.22349988399946596,
-    "gpu_update_seconds": 0.9647970280307759,
-    "producer_batches_per_second": 7.848296043773507,
-    "gpu_batches_per_second": 1.0364874382346263,
-    "producer_headroom": 7.572012698138376,
+    "producer_batch_seconds": 0.06677979871872708,
+    "producer_p95_seconds": 0.21310984199953964,
+    "gpu_update_seconds": 0.9746722100967518,
+    "producer_batches_per_second": 14.974588411264104,
+    "gpu_batches_per_second": 1.0259859567564096,
+    "producer_headroom": 14.595315182095991,
     "sustainable": true,
-    "buffer_seconds": 60.78221276593889,
+    "buffer_seconds": 60.42967702599861,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360152.3332979293,
+    "generation_bytes_per_second": 389274.43597942294,
     "upload_bytes_per_second": 5227697.790039282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.28030680850419,
+    "estimated_full_shards_per_hour": 83.52923211609856,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 624000, "loss": 0.5446088910102844, "lr": 0.0001, "grad_norm": 0.15555424520469768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91337146397, "tokens_per_second": 151456.037639442, "data_wait_s": 1.241100108018145e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624100, "loss": 0.5719287991523743, "lr": 0.0001, "grad_norm": 0.17895209546245056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91351703776, "tokens_per_second": 151328.9869741156, "data_wait_s": 1.020499985315837e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624200, "loss": 0.5563517808914185, "lr": 0.0001, "grad_norm": 0.17141695445517058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91366226013, "tokens_per_second": 150570.36739884503, "data_wait_s": 6.350001058308408e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 624300, "loss": 0.5589475631713867, "lr": 0.0001, "grad_norm": 0.16092953620302844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91380728200, "tokens_per_second": 150561.9621450381, "data_wait_s": 1.2386000889819115e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e1d54ae5e77684e83951888953ab93b58aba65f25dff260ddf9e837baa999d4
-size 422661
+oid sha256:4619c97710adcc1e2866e2833ec630829defc9ff1c9b937af0650728c15da65e
+size 424779