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": 76494.20415156282,
-      "required_pages_per_second_upper_estimate": 0.005292046534212236
+      "required_token_positions_per_second": 76865.01643687322,
+      "required_pages_per_second_upper_estimate": 0.005317700188513052
     },
     {
       "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": 9561.775518945353,
-      "required_pages_per_second_upper_estimate": 0.004003315737204058
+      "required_token_positions_per_second": 9608.127054609153,
+      "required_pages_per_second_upper_estimate": 0.004022722157256359
     },
     {
       "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": 19123.551037890706,
-      "required_pages_per_second_upper_estimate": 0.003982931359237582
+      "required_token_positions_per_second": 19216.254109218306,
+      "required_pages_per_second_upper_estimate": 0.004002238964250729
     },
     {
       "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": 28685.32655683606,
-      "required_pages_per_second_upper_estimate": 0.0017196415271247692
+      "required_token_positions_per_second": 28824.38116382746,
+      "required_pages_per_second_upper_estimate": 0.0017279776384898128
     },
     {
       "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": 28685.32655683606,
-      "required_pages_per_second_upper_estimate": 0.0022301579663768977
+      "required_token_positions_per_second": 28824.38116382746,
+      "required_pages_per_second_upper_estimate": 0.002240968850433903
     },
     {
       "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": 28685.32655683606,
-      "required_pages_per_second_upper_estimate": 0.3679445691670971
+      "required_token_positions_per_second": 28824.38116382746,
+      "required_pages_per_second_upper_estimate": 0.3697282123603784
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0865716622965067,
-    "producer_p95_seconds": 0.2721079570001166,
-    "gpu_update_seconds": 0.9638377288548298,
-    "producer_batches_per_second": 11.551123929849174,
-    "gpu_batches_per_second": 1.0375190450244522,
-    "producer_headroom": 11.133409054266503,
+    "producer_batch_seconds": 0.2460029740543348,
+    "producer_p95_seconds": 0.25977062299716636,
+    "gpu_update_seconds": 0.9591879819677193,
+    "producer_batches_per_second": 4.064991506074758,
+    "gpu_batches_per_second": 1.042548508529639,
+    "producer_headroom": 3.899090999427767,
     "sustainable": true,
-    "buffer_seconds": 60.72177691785428,
+    "buffer_seconds": 60.42884286396632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 538655.0950930445,
+    "generation_bytes_per_second": 487807.4210423075,
     "upload_bytes_per_second": 5590308.97750657,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.58284415811062,
+    "estimated_full_shards_per_hour": 104.67211698009413,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 506400, "loss": 0.5746796727180481, "lr": 0.0003, "grad_norm": 0.15227731313016363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74120257019, "tokens_per_second": 153270.9278461322, "data_wait_s": 1.2759999663103372e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506500, "loss": 0.5777621865272522, "lr": 0.0003, "grad_norm": 0.151934990963674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74135012738, "tokens_per_second": 153225.66126304475, "data_wait_s": 1.8749000446405262e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 506600, "loss": 0.5785301327705383, "lr": 0.0003, "grad_norm": 0.16547915430050822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74149480947, "tokens_per_second": 150849.97201876153, "data_wait_s": 8.549002814106643e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 506700, "loss": 0.5619561076164246, "lr": 0.0003, "grad_norm": 0.1554463108727053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74164155673, "tokens_per_second": 152544.85657892728, "data_wait_s": 1.4468998415395617e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:319fe383fc75434a6c542c4383a5b35c3a0d545d6459006e307f5b7e02d09521
-size 460785
+oid sha256:d5a9d4ea3e62bf45d974ba6f2ce7ebe447a2b7aabd59e7e171f1ec4c85e4166d
+size 462903