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": 76723.49785940888,
-      "required_pages_per_second_upper_estimate": 0.005307909604955728
+      "required_token_positions_per_second": 76560.5757908421,
+      "required_pages_per_second_upper_estimate": 0.005296638278220997
     },
     {
       "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": 9590.43723242611,
-      "required_pages_per_second_upper_estimate": 0.004015315798113812
+      "required_token_positions_per_second": 9570.071973855262,
+      "required_pages_per_second_upper_estimate": 0.004006789289625157
     },
     {
       "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": 19180.87446485222,
-      "required_pages_per_second_upper_estimate": 0.003994870317353238
+      "required_token_positions_per_second": 19140.143947710523,
+      "required_pages_per_second_upper_estimate": 0.003986387224768592
     },
     {
       "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": 28771.31169727833,
-      "required_pages_per_second_upper_estimate": 0.001724796205002576
+      "required_token_positions_per_second": 28710.21592156578,
+      "required_pages_per_second_upper_estimate": 0.0017211336065364566
     },
     {
       "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": 28771.31169727833,
-      "required_pages_per_second_upper_estimate": 0.002236842932837622
+      "required_token_positions_per_second": 28710.21592156578,
+      "required_pages_per_second_upper_estimate": 0.0022320930050078883
     },
     {
       "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": 28771.31169727833,
-      "required_pages_per_second_upper_estimate": 0.3690474942250398
+      "required_token_positions_per_second": 28710.21592156578,
+      "required_pages_per_second_upper_estimate": 0.36826382321373224
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08475637591419627,
-    "producer_p95_seconds": 0.2634155939995253,
-    "gpu_update_seconds": 0.9609572302751637,
-    "producer_batches_per_second": 11.798522402755365,
-    "gpu_batches_per_second": 1.0406290399767915,
-    "producer_headroom": 11.337875409491266,
+    "producer_batch_seconds": 0.10014666849215814,
+    "producer_p95_seconds": 0.2635770950000733,
+    "gpu_update_seconds": 0.9630021618622556,
+    "producer_batches_per_second": 9.985354630926178,
+    "gpu_batches_per_second": 1.0384192679964475,
+    "producer_headroom": 9.615918096543194,
     "sustainable": true,
-    "buffer_seconds": 60.54030550733531,
+    "buffer_seconds": 60.6691361973221,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305626.9061350543,
-    "upload_bytes_per_second": 3594470.5342969554,
+    "generation_bytes_per_second": 306395.6061295517,
+    "upload_bytes_per_second": 3547244.6910242583,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.58041942633363,
+    "estimated_full_shards_per_hour": 65.74536455073275,
     "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": 1.1171746460004215,
-      "bytes": 6111968,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2683859379994828,
       "bytes": 7618531,
@@ -562,6 +556,12 @@
       "bytes": 16470150,
       "pages": 45,
       "shards": 4
+    },
+    {
+      "seconds": 1.350964933000796,
+      "bytes": 1228633,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -192,3 +192,4 @@
 {"step": 861700, "loss": 0.49686741828918457, "lr": 3e-05, "grad_norm": 0.304775112780428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126128248430, "tokens_per_second": 151406.8000760035, "data_wait_s": 1.575900023453869e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861800, "loss": 0.4809182584285736, "lr": 3e-05, "grad_norm": 0.3109340984030623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126142892141, "tokens_per_second": 151753.0956627732, "data_wait_s": 1.397199957864359e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861900, "loss": 0.5383116602897644, "lr": 3e-05, "grad_norm": 0.31303202220947224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126157505547, "tokens_per_second": 151777.9644042938, "data_wait_s": 1.0981002560583875e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 862000, "loss": 0.5001435875892639, "lr": 3e-05, "grad_norm": 0.3377530224964609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126172153606, "tokens_per_second": 151653.17379325684, "data_wait_s": 1.5920999430818483e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9853d59a2598f97ad17d6be0eafd5d11ae0c2df86f9452c46cee0c90d791f76d
-size 414189
+oid sha256:46b5c99db6a01c31363c5f2bcf67abbbbf0d18861ff78aa7542ba43e709256b1
+size 416307