Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75223.90261595906,
-      "required_pages_per_second_upper_estimate": 0.006520217225513188
+      "required_token_positions_per_second": 75999.12812604562,
+      "required_pages_per_second_upper_estimate": 0.006587411807936398
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9402.987826994882,
-      "required_pages_per_second_upper_estimate": 0.0037509924313845868
+      "required_token_positions_per_second": 9499.891015755702,
+      "required_pages_per_second_upper_estimate": 0.0037896485622130613
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18805.975653989764,
-      "required_pages_per_second_upper_estimate": 0.004350431493683183
+      "required_token_positions_per_second": 18999.782031511404,
+      "required_pages_per_second_upper_estimate": 0.004395265188247066
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28208.963480984647,
-      "required_pages_per_second_upper_estimate": 0.0017177778749986998
+      "required_token_positions_per_second": 28499.673047267104,
+      "required_pages_per_second_upper_estimate": 0.0017354805623500836
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28208.963480984647,
-      "required_pages_per_second_upper_estimate": 0.00217160409617493
+      "required_token_positions_per_second": 28499.673047267104,
+      "required_pages_per_second_upper_estimate": 0.0021939837233229385
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28208.963480984647,
-      "required_pages_per_second_upper_estimate": 0.35026091710623247
+      "required_token_positions_per_second": 28499.673047267104,
+      "required_pages_per_second_upper_estimate": 0.3538705569771298
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09913053871886746,
-    "producer_p95_seconds": 0.2943986450009106,
-    "gpu_update_seconds": 0.9801139988230058,
-    "producer_batches_per_second": 10.087708721486758,
-    "gpu_batches_per_second": 1.0202894777555211,
-    "producer_headroom": 9.887104533978098,
+    "producer_batch_seconds": 0.12395837693742351,
+    "producer_p95_seconds": 0.32236783699772786,
+    "gpu_update_seconds": 0.970116392358095,
+    "producer_batches_per_second": 8.067224052996584,
+    "gpu_batches_per_second": 1.0308041466748807,
+    "producer_headroom": 7.826146294637495,
     "sustainable": true,
-    "buffer_seconds": 60.76706792702636,
+    "buffer_seconds": 60.14721632620189,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465760.3441272158,
+    "generation_bytes_per_second": 466257.7871084776,
     "upload_bytes_per_second": 5728821.4302538335,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.94132750379902,
+    "estimated_full_shards_per_hour": 100.04806718769785,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -247,3 +247,4 @@
 {"step": 24700, "loss": 0.6467013955116272, "lr": 0.0003, "grad_norm": 0.13487899901526593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3614016837, "tokens_per_second": 146990.5511808094, "data_wait_s": 1.1520001862663776e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 24800, "loss": 0.6468101106584072, "lr": 0.0003, "grad_norm": 0.13603707268163948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3628643746, "tokens_per_second": 148611.72245921195, "data_wait_s": 9.395000233780593e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 24900, "loss": 0.7069809772074223, "lr": 0.0003, "grad_norm": 0.16682850327811186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3643309839, "tokens_per_second": 148773.50165499502, "data_wait_s": 1.2054999388055876e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 25000, "loss": 0.7111044470220804, "lr": 0.0003, "grad_norm": 0.1274851893123072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3657994460, "tokens_per_second": 150374.07818280114, "data_wait_s": 1.0256997484248132e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5185b13090e06629f3989e751ca30db66fe3edaf37a110c6abdf192a80c97bef
-size 524905
+oid sha256:9ceb7995fd69aa1a6e540e0cb038f737644fc90e0a60370d3acbfd2bc45f494f
+size 527023