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": 77333.89235156549,
-      "required_pages_per_second_upper_estimate": 0.005350138112233474
+      "required_token_positions_per_second": 77043.90787572875,
+      "required_pages_per_second_upper_estimate": 0.005330076313338403
     },
     {
       "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": 9666.736543945686,
-      "required_pages_per_second_upper_estimate": 0.0040472607265362535
+      "required_token_positions_per_second": 9630.488484466094,
+      "required_pages_per_second_upper_estimate": 0.004032084420977705
     },
     {
       "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": 19333.473087891372,
-      "required_pages_per_second_upper_estimate": 0.004026652586235982
+      "required_token_positions_per_second": 19260.976968932187,
+      "required_pages_per_second_upper_estimate": 0.0040115535565080065
     },
     {
       "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": 29000.20963183706,
-      "required_pages_per_second_upper_estimate": 0.001738518286672463
+      "required_token_positions_per_second": 28891.465453398283,
+      "required_pages_per_second_upper_estimate": 0.0017319992392178045
     },
     {
       "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": 29000.20963183706,
-      "required_pages_per_second_upper_estimate": 0.0022546387404339526
+      "required_token_positions_per_second": 28891.465453398283,
+      "required_pages_per_second_upper_estimate": 0.002246184358875429
     },
     {
       "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": 29000.20963183706,
-      "required_pages_per_second_upper_estimate": 0.37198355115810544
+      "required_token_positions_per_second": 28891.465453398283,
+      "required_pages_per_second_upper_estimate": 0.3705886975974947
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4869148282421918,
-    "producer_p95_seconds": 0.22407763799856184,
-    "gpu_update_seconds": 0.9533724187168435,
-    "producer_batches_per_second": 2.053747271591818,
-    "gpu_batches_per_second": 1.0489080451329955,
-    "producer_headroom": 1.9579860037506096,
+    "producer_batch_seconds": 0.08977657719512422,
+    "producer_p95_seconds": 0.22217336600078852,
+    "gpu_update_seconds": 0.9569608036877194,
+    "producer_batches_per_second": 11.138762818129695,
+    "gpu_batches_per_second": 1.0449748789568245,
+    "producer_headroom": 10.65935941852428,
     "sustainable": true,
-    "buffer_seconds": 60.06246237916114,
+    "buffer_seconds": 60.28853063232632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301798.2087931432,
+    "generation_bytes_per_second": 312662.31897651224,
     "upload_bytes_per_second": 4776004.706385802,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.7588701042721,
+    "estimated_full_shards_per_hour": 67.09005524608159,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 944300, "loss": 0.48684871196746826, "lr": 3e-05, "grad_norm": 0.33078584826771573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138219204026, "tokens_per_second": 150047.56068361085, "data_wait_s": 1.2052998499711975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944400, "loss": 0.5279709100723267, "lr": 3e-05, "grad_norm": 0.32609924167972903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138233576903, "tokens_per_second": 144423.03887124828, "data_wait_s": 1.3274002412799746e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944500, "loss": 0.577720582485199, "lr": 3e-05, "grad_norm": 0.35585286234893304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138248100173, "tokens_per_second": 150613.3662130495, "data_wait_s": 1.7441001546103507e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 944600, "loss": 0.555435836315155, "lr": 3e-05, "grad_norm": 0.35039679697845094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138262793802, "tokens_per_second": 153374.08155014302, "data_wait_s": 1.309999788645655e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dccfee78f62bd553455e586d7536ba2c3b8dfaea81d6b34cc845f2f0a9c7ecf2
-size 415524
+oid sha256:965ed443d95804a62bd5001dd96be135c7857d0e8f57b08e05e4ee2ef5199254
+size 417642