Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75579.03941565944,
-      "required_pages_per_second_upper_estimate": 0.006550999583225227
+      "required_token_positions_per_second": 75492.62586324968,
+      "required_pages_per_second_upper_estimate": 0.0065435094755154945
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9447.37992695743,
-      "required_pages_per_second_upper_estimate": 0.0037687011037806886
+      "required_token_positions_per_second": 9436.57823290621,
+      "required_pages_per_second_upper_estimate": 0.0037643921465239385
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18894.75985391486,
-      "required_pages_per_second_upper_estimate": 0.004370970155787303
+      "required_token_positions_per_second": 18873.15646581242,
+      "required_pages_per_second_upper_estimate": 0.004365972592156448
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28342.13978087229,
-      "required_pages_per_second_upper_estimate": 0.0017258876129398117
+      "required_token_positions_per_second": 28309.734698718632,
+      "required_pages_per_second_upper_estimate": 0.0017239143134529806
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28342.13978087229,
-      "required_pages_per_second_upper_estimate": 0.002181856376395874
+      "required_token_positions_per_second": 28309.734698718632,
+      "required_pages_per_second_upper_estimate": 0.0021793617434686767
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28342.13978087229,
-      "required_pages_per_second_upper_estimate": 0.3519145210384331
+      "required_token_positions_per_second": 28309.734698718632,
+      "required_pages_per_second_upper_estimate": 0.3515121583709181
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13585464850791595,
-    "producer_p95_seconds": 0.4217981690017041,
-    "gpu_update_seconds": 0.9755085612364118,
-    "producer_batches_per_second": 7.360808120906751,
-    "gpu_batches_per_second": 1.0251063288799294,
-    "producer_headroom": 7.180531339563041,
+    "producer_batch_seconds": 0.1185122089922288,
+    "producer_p95_seconds": 0.39575560000230325,
+    "gpu_update_seconds": 0.9766251889761234,
+    "producer_batches_per_second": 8.437949207963653,
+    "gpu_batches_per_second": 1.0239342700636078,
+    "producer_headroom": 8.240713739798434,
     "sustainable": true,
-    "buffer_seconds": 60.48153079665753,
+    "buffer_seconds": 60.55076171651965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381919.22781254374,
+    "generation_bytes_per_second": 382873.19731244753,
     "upload_bytes_per_second": 3782178.175520555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.95097566396937,
+    "estimated_full_shards_per_hour": 82.15567531137533,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -384,3 +384,4 @@
 {"step": 38400, "loss": 0.6338763236999512, "lr": 0.0003, "grad_norm": 0.13404358026008098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5619227380, "tokens_per_second": 147564.20487816597, "data_wait_s": 9.408002370037138e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 38500, "loss": 0.6896477043628693, "lr": 0.0003, "grad_norm": 0.1378860124558519, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5633848010, "tokens_per_second": 148474.49969914253, "data_wait_s": 1.6270001651719213e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 38600, "loss": 0.6759868860244751, "lr": 0.0003, "grad_norm": 0.12844563084328195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5648475620, "tokens_per_second": 148436.1762926683, "data_wait_s": 1.2041004083584994e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 38700, "loss": 0.6552939936518669, "lr": 0.0003, "grad_norm": 0.12835237871211294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5663203830, "tokens_per_second": 149862.03059944644, "data_wait_s": 9.670999133959413e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e978639d4d67293759cfe6d048650096278ecfb755955e88d8fb5290f73347e7
-size 817420
+oid sha256:7c8438a7c2e575ac1c7ea4d79da87fada9c7cb75f30da19565b75ebaed7e9d30
+size 819538