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": 76842.78916151651,
-      "required_pages_per_second_upper_estimate": 0.00531616245402949
+      "required_token_positions_per_second": 76455.55900804087,
+      "required_pages_per_second_upper_estimate": 0.005289372973514301
     },
     {
       "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": 9605.348645189564,
-      "required_pages_per_second_upper_estimate": 0.004021558895252164
+      "required_token_positions_per_second": 9556.94487600511,
+      "required_pages_per_second_upper_estimate": 0.004001293247880273
     },
     {
       "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": 19210.69729037913,
-      "required_pages_per_second_upper_estimate": 0.004001081625429695
+      "required_token_positions_per_second": 19113.88975201022,
+      "required_pages_per_second_upper_estimate": 0.003980919168173919
     },
     {
       "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": 28816.045935568694,
-      "required_pages_per_second_upper_estimate": 0.0017274779542828566
+      "required_token_positions_per_second": 28670.834628015327,
+      "required_pages_per_second_upper_estimate": 0.0017187727581198604
     },
     {
       "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": 28816.045935568694,
-      "required_pages_per_second_upper_estimate": 0.002240320823099578
+      "required_token_positions_per_second": 28670.834628015327,
+      "required_pages_per_second_upper_estimate": 0.002229031282654344
     },
     {
       "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": 28816.045935568694,
-      "required_pages_per_second_upper_estimate": 0.3696212970019458
+      "required_token_positions_per_second": 28670.834628015327,
+      "required_pages_per_second_upper_estimate": 0.3677586822644056
     }
   ],
   "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": 0.08888798501601514,
-    "producer_p95_seconds": 0.32414882900047814,
-    "gpu_update_seconds": 0.959465433315161,
-    "producer_batches_per_second": 11.250114397573844,
-    "gpu_batches_per_second": 1.0422470318131039,
-    "producer_headroom": 10.79409588531332,
+    "producer_batch_seconds": 0.10894605182801342,
+    "producer_p95_seconds": 0.34728073600126663,
+    "gpu_update_seconds": 0.9643249092227026,
+    "producer_batches_per_second": 9.178854884788665,
+    "gpu_batches_per_second": 1.0369948867193044,
+    "producer_headroom": 8.85139840354219,
     "sustainable": true,
-    "buffer_seconds": 60.446322298855144,
+    "buffer_seconds": 60.752469281030265,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396435.2650906284,
-    "upload_bytes_per_second": 4800066.77100214,
+    "generation_bytes_per_second": 396931.5223456809,
+    "upload_bytes_per_second": 4786988.736817365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.065779347793,
+    "estimated_full_shards_per_hour": 85.17226460244962,
     "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": 1.191631180998229,
-      "bytes": 7251240,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1317712890013354,
       "bytes": 6287146,
@@ -562,6 +556,12 @@
       "bytes": 11302182,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.1526390029976028,
+      "bytes": 5901438,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 961200, "loss": 0.5192878246307373, "lr": 3e-05, "grad_norm": 0.35351720056532393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140693943165, "tokens_per_second": 154038.12617002902, "data_wait_s": 1.1238997103646398e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961300, "loss": 0.5140148997306824, "lr": 3e-05, "grad_norm": 0.35353837122117165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140708527642, "tokens_per_second": 151693.61127884014, "data_wait_s": 1.3297998521011323e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961400, "loss": 0.4920903146266937, "lr": 3e-05, "grad_norm": 0.3526757492499405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140723051873, "tokens_per_second": 150683.633431203, "data_wait_s": 1.4518998796120286e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 961500, "loss": 0.5077930688858032, "lr": 3e-05, "grad_norm": 0.3514096955737956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140737578155, "tokens_per_second": 150925.29597307602, "data_wait_s": 1.22560013551265e-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 sha256:7bc539922d19a62aac372f6ba52cef073d8968b0a920c79cb03bff26734f6f88
-size 776598
+oid sha256:0cac8fb9fc95791e58dccaea1b7d4bacefa8fab082bf99dc3767968ac81c3985
+size 778716