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": 76216.00490696738,
-      "required_pages_per_second_upper_estimate": 0.005272800064959942
+      "required_token_positions_per_second": 76674.12350258778,
+      "required_pages_per_second_upper_estimate": 0.0053044937723865536
     },
     {
       "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": 9527.000613370923,
-      "required_pages_per_second_upper_estimate": 0.003988756210422649
+      "required_token_positions_per_second": 9584.265437823473,
+      "required_pages_per_second_upper_estimate": 0.004012731796595416
     },
     {
       "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": 19054.001226741846,
-      "required_pages_per_second_upper_estimate": 0.003968445967727136
+      "required_token_positions_per_second": 19168.530875646946,
+      "required_pages_per_second_upper_estimate": 0.003992299473244117
     },
     {
       "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": 28581.00184011277,
-      "required_pages_per_second_upper_estimate": 0.0017133874196518272
+      "required_token_positions_per_second": 28752.79631347042,
+      "required_pages_per_second_upper_estimate": 0.0017236862360146459
     },
     {
       "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": 28581.00184011277,
-      "required_pages_per_second_upper_estimate": 0.0022220471785275955
+      "required_token_positions_per_second": 28752.79631347042,
+      "required_pages_per_second_upper_estimate": 0.0022354034431871232
     },
     {
       "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": 28581.00184011277,
-      "required_pages_per_second_upper_estimate": 0.36660640371612435
+      "required_token_positions_per_second": 28752.79631347042,
+      "required_pages_per_second_upper_estimate": 0.3688099987618222
     }
   ],
   "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": 0.07657060054680187,
-    "producer_p95_seconds": 0.19895318199996836,
-    "gpu_update_seconds": 0.9673558734808475,
-    "producer_batches_per_second": 13.059842718469667,
-    "gpu_batches_per_second": 1.033745726277227,
-    "producer_headroom": 12.633515560447712,
+    "producer_batch_seconds": 0.09047906085936575,
+    "producer_p95_seconds": 0.20201538699984667,
+    "gpu_update_seconds": 0.9615760393728094,
+    "producer_batches_per_second": 11.052280942154443,
+    "gpu_batches_per_second": 1.0399593574027206,
+    "producer_headroom": 10.627608534392452,
     "sustainable": true,
-    "buffer_seconds": 60.94342002929339,
+    "buffer_seconds": 60.579290480486996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280012.9714006981,
-    "upload_bytes_per_second": 4905477.534129387,
+    "generation_bytes_per_second": 305585.7295659716,
+    "upload_bytes_per_second": 4900843.344162811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.084265294224814,
+    "estimated_full_shards_per_hour": 65.57158389314995,
     "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": 1.2438676930000838,
-      "bytes": 6205549,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.3983366309998928,
       "bytes": 1085324,
@@ -562,6 +556,12 @@
       "bytes": 6603515,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.285518773000149,
+      "bytes": 6002809,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -95,3 +95,4 @@
 {"step": 852000, "loss": 0.5257346034049988, "lr": 3e-05, "grad_norm": 0.30198060240331964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124709130689, "tokens_per_second": 144589.06245153135, "data_wait_s": 1.0116998964804225e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852100, "loss": 0.5304078459739685, "lr": 3e-05, "grad_norm": 0.30842763203355267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124723926557, "tokens_per_second": 153995.97442909091, "data_wait_s": 1.355600034003146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852200, "loss": 0.5567572116851807, "lr": 3e-05, "grad_norm": 0.29481274002551955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124738572804, "tokens_per_second": 152241.31199595297, "data_wait_s": 1.2762000551447272e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 852300, "loss": 0.5323869585990906, "lr": 3e-05, "grad_norm": 0.28288396236856206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124753180994, "tokens_per_second": 152524.21277255262, "data_wait_s": 1.3941998986410908e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f085c2adf621f218fcf6f50c4f465d35eda3b6e7520aaccc7f5335cadd188ae
-size 207177
+oid sha256:7a4a4e67ae0bf68d24a77156391ba6e9583a4f26c7da89cd7c333b67415495ce
+size 209295