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": 77634.8271644069,
-      "required_pages_per_second_upper_estimate": 0.005370957480850811
+      "required_token_positions_per_second": 77610.41667237945,
+      "required_pages_per_second_upper_estimate": 0.005369268706372207
     },
     {
       "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": 9704.353395550863,
-      "required_pages_per_second_upper_estimate": 0.004063010116774154
+      "required_token_positions_per_second": 9701.302084047431,
+      "required_pages_per_second_upper_estimate": 0.004061732596366297
     },
     {
       "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": 19408.706791101726,
-      "required_pages_per_second_upper_estimate": 0.004042321782568524
+      "required_token_positions_per_second": 19402.604168094862,
+      "required_pages_per_second_upper_estimate": 0.004041050767133185
     },
     {
       "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": 29113.06018665259,
-      "required_pages_per_second_upper_estimate": 0.0017452835051208339
+      "required_token_positions_per_second": 29103.906252142297,
+      "required_pages_per_second_upper_estimate": 0.0017447347407241905
     },
     {
       "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": 29113.06018665259,
-      "required_pages_per_second_upper_estimate": 0.002263412374693728
+      "required_token_positions_per_second": 29103.906252142297,
+      "required_pages_per_second_upper_estimate": 0.0022627006965494556
     },
     {
       "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": 29113.06018665259,
-      "required_pages_per_second_upper_estimate": 0.3734310769057938
+      "required_token_positions_per_second": 29103.906252142297,
+      "required_pages_per_second_upper_estimate": 0.37331366006262484
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06189826855468539,
-    "producer_p95_seconds": 0.1497266540000055,
-    "gpu_update_seconds": 0.9496768743217082,
-    "producer_batches_per_second": 16.155540750166995,
-    "gpu_batches_per_second": 1.0529897347602934,
-    "producer_headroom": 15.342543442595575,
+    "producer_batch_seconds": 0.0724234139374964,
+    "producer_p95_seconds": 0.15600951599981272,
+    "gpu_update_seconds": 0.9499755723671929,
+    "producer_batches_per_second": 13.807689331837217,
+    "gpu_batches_per_second": 1.0526586462725078,
+    "producer_headroom": 13.116967576080445,
     "sustainable": true,
-    "buffer_seconds": 60.779319956589326,
+    "buffer_seconds": 60.798436631500344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444301.7669219209,
+    "generation_bytes_per_second": 442778.06118253944,
     "upload_bytes_per_second": 4904212.914559329,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.33681636565419,
+    "estimated_full_shards_per_hour": 95.00986458403719,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -27,3 +27,4 @@
 {"step": 887700, "loss": 0.5393361449241638, "lr": 3e-05, "grad_norm": 0.3169755863280241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129934846871, "tokens_per_second": 154480.24010580132, "data_wait_s": 1.3175000276532955e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887800, "loss": 0.5599460601806641, "lr": 3e-05, "grad_norm": 0.32537648868909574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129949566163, "tokens_per_second": 153708.03141012922, "data_wait_s": 2.7463000151328743e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887900, "loss": 0.5958407521247864, "lr": 3e-05, "grad_norm": 0.33544832627131727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129964000753, "tokens_per_second": 150912.60629945368, "data_wait_s": 1.459299983253004e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 888000, "loss": 0.49947598576545715, "lr": 3e-05, "grad_norm": 0.3130021412299721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129978566661, "tokens_per_second": 151972.8083974794, "data_wait_s": 1.097999984267517e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2de9c6c5a84fe0109c4904879915ce7fb3f5c7eda160775c2aaf70025d9de732
-size 61587
+oid sha256:1f8b9d8f7e5f9d49c9752ff0b2a6f52c81ffb0d9ddaba4fd00e1f12ae91f69a5
+size 63705