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": 75999.12812604562,
-      "required_pages_per_second_upper_estimate": 0.006587411807936398
+      "required_token_positions_per_second": 75601.9963858874,
+      "required_pages_per_second_upper_estimate": 0.006552989435220677
     },
     {
       "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": 9499.891015755702,
-      "required_pages_per_second_upper_estimate": 0.0037896485622130613
+      "required_token_positions_per_second": 9450.249548235925,
+      "required_pages_per_second_upper_estimate": 0.0037698458386133415
     },
     {
       "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": 18999.782031511404,
-      "required_pages_per_second_upper_estimate": 0.004395265188247066
+      "required_token_positions_per_second": 18900.49909647185,
+      "required_pages_per_second_upper_estimate": 0.004372297828545643
     },
     {
       "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": 28499.673047267104,
-      "required_pages_per_second_upper_estimate": 0.0017354805623500836
+      "required_token_positions_per_second": 28350.74864470777,
+      "required_pages_per_second_upper_estimate": 0.0017264118475801748
     },
     {
       "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": 28499.673047267104,
-      "required_pages_per_second_upper_estimate": 0.0021939837233229385
+      "required_token_positions_per_second": 28350.74864470777,
+      "required_pages_per_second_upper_estimate": 0.0021825191105647895
     },
     {
       "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": 28499.673047267104,
-      "required_pages_per_second_upper_estimate": 0.3538705569771298
+      "required_token_positions_per_second": 28350.74864470777,
+      "required_pages_per_second_upper_estimate": 0.35202141431525596
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.27939625888305386,
+    "producer_p95_seconds": 0.3554270029999316,
+    "gpu_update_seconds": 0.9752123425905032,
+    "producer_batches_per_second": 3.5791459914234833,
+    "gpu_batches_per_second": 1.0254177027165716,
+    "producer_headroom": 3.490427346769504,
     "sustainable": true,
-    "buffer_seconds": 60.14721632620189,
+    "buffer_seconds": 60.4631652406112,
     "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": 466257.7871084776,
-    "upload_bytes_per_second": 5728821.4302538335,
+    "generation_bytes_per_second": 533483.29474944,
+    "upload_bytes_per_second": 5678488.158291832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.04806718769785,
+    "estimated_full_shards_per_hour": 114.4730961977234,
     "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": 1.0224713220013655,
-      "bytes": 6909652,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.018912650000857,
       "bytes": 5819747,
@@ -562,6 +556,12 @@
       "bytes": 5783909,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8641120300017064,
+      "bytes": 2329685,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -248,3 +248,4 @@
 {"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}
+{"step": 25100, "loss": 0.6755372695624828, "lr": 0.0003, "grad_norm": 0.13377661139797237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3672698039, "tokens_per_second": 138098.96691315155, "data_wait_s": 9.449999197386205e-06, "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 sha256:82cb99715bc3db050b5131cc58818b50f585e6d8d26288b06782b6dd3af50f8b
-size 527806
+oid sha256:713b43497834f56be2f2bef8065041cd2140e9e4ac0956a13ff4368fdc2548fd
+size 529924