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": 75419.27635683188,
-      "required_pages_per_second_upper_estimate": 0.006537151726201824
+      "required_token_positions_per_second": 75157.8084090762,
+      "required_pages_per_second_upper_estimate": 0.006514488347174817
     },
     {
       "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": 9427.409544603985,
-      "required_pages_per_second_upper_estimate": 0.0037607346196760747
+      "required_token_positions_per_second": 9394.726051134525,
+      "required_pages_per_second_upper_estimate": 0.0037476966854693332
     },
     {
       "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": 18854.81908920797,
-      "required_pages_per_second_upper_estimate": 0.004361730562805814
+      "required_token_positions_per_second": 18789.45210226905,
+      "required_pages_per_second_upper_estimate": 0.004346609060797173
     },
     {
       "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": 28282.228633811956,
-      "required_pages_per_second_upper_estimate": 0.0017222393384133334
+      "required_token_positions_per_second": 28184.178153403573,
+      "required_pages_per_second_upper_estimate": 0.0017162685785875762
     },
     {
       "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": 28282.228633811956,
-      "required_pages_per_second_upper_estimate": 0.0021772442504505
+      "required_token_positions_per_second": 28184.178153403573,
+      "required_pages_per_second_upper_estimate": 0.0021696960530475607
     },
     {
       "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": 28282.228633811956,
-      "required_pages_per_second_upper_estimate": 0.3511706251016546
+      "required_token_positions_per_second": 28184.178153403573,
+      "required_pages_per_second_upper_estimate": 0.3499531662888309
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.05029131468747394,
-    "producer_p95_seconds": 0.1713132200002292,
-    "gpu_update_seconds": 0.9775750121384097,
-    "producer_batches_per_second": 19.88414910634798,
-    "gpu_batches_per_second": 1.02293940371137,
-    "producer_headroom": 19.438247304000075,
+    "producer_batch_seconds": 0.1424565040703314,
+    "producer_p95_seconds": 0.17702674200063484,
+    "gpu_update_seconds": 0.980975916683282,
+    "producer_batches_per_second": 7.019686510812421,
+    "gpu_batches_per_second": 1.0193930177012287,
+    "producer_headroom": 6.886143409773484,
     "sustainable": true,
-    "buffer_seconds": 60.6096507525814,
+    "buffer_seconds": 60.820506834363485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272400.64527585835,
-    "upload_bytes_per_second": 5093472.592298259,
+    "generation_bytes_per_second": 294320.86622935033,
+    "upload_bytes_per_second": 5227156.164053405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.45083731371701,
+    "estimated_full_shards_per_hour": 63.15440645370848,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 1203088,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.1118799440000657,
+      "bytes": 10497270,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 318400, "loss": 0.6080475449562073, "lr": 0.0003, "grad_norm": 0.1703554844603877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46605145498, "tokens_per_second": 148761.08692721464, "data_wait_s": 1.060699969457346e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 318500, "loss": 0.605508603155613, "lr": 0.0003, "grad_norm": 0.14967705484341146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46619811868, "tokens_per_second": 148610.7748379242, "data_wait_s": 1.3629000022774562e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 318600, "loss": 0.6292090825736523, "lr": 0.0003, "grad_norm": 0.16122556275773078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46634337336, "tokens_per_second": 147293.07660231058, "data_wait_s": 1.529099972685799e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 318700, "loss": 0.5911545790731907, "lr": 0.0003, "grad_norm": 0.16510909979838936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46649008923, "tokens_per_second": 147937.58094382394, "data_wait_s": 1.1021000318578444e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:36260a2378439f4fdcfeaeaee89a0c1cc3eae9f37856a07119418bfcbd715ecd
-size 76413
+oid sha256:3d753e52a3a8280c2c834b251b2e2944dca5b8b89a1ece1b42fad68d06a85bbf
+size 78531