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": 76883.84360296112,
-      "required_pages_per_second_upper_estimate": 0.005319002695548056
+      "required_token_positions_per_second": 76540.65722640579,
+      "required_pages_per_second_upper_estimate": 0.005295260265715857
     },
     {
       "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": 9610.48045037014,
-      "required_pages_per_second_upper_estimate": 0.0040237074749169925
+      "required_token_positions_per_second": 9567.582153300724,
+      "required_pages_per_second_upper_estimate": 0.004005746853752338
     },
     {
       "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": 19220.96090074028,
-      "required_pages_per_second_upper_estimate": 0.004003219264798316
+      "required_token_positions_per_second": 19135.164306601448,
+      "required_pages_per_second_upper_estimate": 0.003985350096847539
     },
     {
       "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": 28831.441351110418,
-      "required_pages_per_second_upper_estimate": 0.0017284008859371443
+      "required_token_positions_per_second": 28702.746459902173,
+      "required_pages_per_second_upper_estimate": 0.001720685823714931
     },
     {
       "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": 28831.441351110418,
-      "required_pages_per_second_upper_estimate": 0.002241517748940671
+      "required_token_positions_per_second": 28702.746459902173,
+      "required_pages_per_second_upper_estimate": 0.0022315122872182324
     },
     {
       "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": 28831.441351110418,
-      "required_pages_per_second_upper_estimate": 0.3698187728622057
+      "required_token_positions_per_second": 28702.746459902173,
+      "required_pages_per_second_upper_estimate": 0.36816801297959456
     }
   ],
   "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": 0.08387707603907302,
-    "producer_p95_seconds": 0.23393648199999006,
-    "gpu_update_seconds": 0.9589530978802474,
-    "producer_batches_per_second": 11.922208632238961,
-    "gpu_batches_per_second": 1.042803868312732,
-    "producer_headroom": 11.432838901460178,
+    "producer_batch_seconds": 0.9605098017265732,
+    "producer_p95_seconds": 0.27675607800119906,
+    "gpu_update_seconds": 0.9632527688116655,
+    "producer_batches_per_second": 1.0411137899919822,
+    "gpu_batches_per_second": 1.0381491051758598,
+    "producer_headroom": 1.0028557408577838,
     "sustainable": true,
-    "buffer_seconds": 60.41404516645559,
+    "buffer_seconds": 60.68492443513493,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267865.21261734137,
-    "upload_bytes_per_second": 3925480.963442974,
+    "generation_bytes_per_second": 235278.47203880607,
+    "upload_bytes_per_second": 3940447.898117824,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.47763904466802,
+    "estimated_full_shards_per_hour": 50.485283097010964,
     "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": 1.4180366079999658,
-      "bytes": 818969,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.935665021999739,
       "bytes": 7829235,
@@ -562,6 +556,12 @@
       "bytes": 5896707,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.9101363129993842,
+      "bytes": 4340262,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -202,3 +202,4 @@
 {"step": 820200, "loss": 0.5426285862922668, "lr": 3e-05, "grad_norm": 0.2829531499152139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120054682315, "tokens_per_second": 152700.30204450135, "data_wait_s": 5.71200143895112e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 820300, "loss": 0.511267364025116, "lr": 3e-05, "grad_norm": 0.2831174499490357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120069411306, "tokens_per_second": 152653.14703792115, "data_wait_s": 9.655999747337773e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 820400, "loss": 0.49906840920448303, "lr": 3e-05, "grad_norm": 0.2801640626981123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120084012610, "tokens_per_second": 118963.99736858156, "data_wait_s": 1.4907000149833038e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 820500, "loss": 0.5685487985610962, "lr": 3e-05, "grad_norm": 0.292315157138596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120098478571, "tokens_per_second": 149555.9637116822, "data_wait_s": 1.030999919748865e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd8f381d1ddaecb273993cf50374e807c7c83d4399863bec50a19737ad975b82
-size 435369
+oid sha256:e9d1a44ab4eaffd797db016015892b21936be9c06b2b9404f03f307e75d1107e
+size 437487