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": 75887.5622437391,
-      "required_pages_per_second_upper_estimate": 0.006577741560019213
+      "required_token_positions_per_second": 75690.46541954389,
+      "required_pages_per_second_upper_estimate": 0.0065606577068353066
     },
     {
       "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": 9485.945280467387,
-      "required_pages_per_second_upper_estimate": 0.003784085399899229
+      "required_token_positions_per_second": 9461.308177442987,
+      "required_pages_per_second_upper_estimate": 0.0037742572911452794
     },
     {
       "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": 18971.890560934775,
-      "required_pages_per_second_upper_estimate": 0.0043888129874022795
+      "required_token_positions_per_second": 18922.616354885973,
+      "required_pages_per_second_upper_estimate": 0.004377414267029297
     },
     {
       "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": 28457.83584140216,
-      "required_pages_per_second_upper_estimate": 0.0017329328960157635
+      "required_token_positions_per_second": 28383.924532328958,
+      "required_pages_per_second_upper_estimate": 0.0017284320850758747
     },
     {
       "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": 28457.83584140216,
-      "required_pages_per_second_upper_estimate": 0.002190762979402657
+      "required_token_positions_per_second": 28383.924532328958,
+      "required_pages_per_second_upper_estimate": 0.0021850730822306035
     },
     {
       "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": 28457.83584140216,
-      "required_pages_per_second_upper_estimate": 0.353351078900408
+      "required_token_positions_per_second": 28383.924532328958,
+      "required_pages_per_second_upper_estimate": 0.3524333478069578
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10466230157032896,
-    "producer_p95_seconds": 0.2020371630005684,
-    "gpu_update_seconds": 0.9715426061941097,
-    "producer_batches_per_second": 9.554538596956414,
-    "gpu_batches_per_second": 1.0292909375507147,
-    "producer_headroom": 9.282641329469246,
+    "producer_batch_seconds": 0.06372208896090115,
+    "producer_p95_seconds": 0.21399366199875658,
+    "gpu_update_seconds": 0.974072488408333,
+    "producer_batches_per_second": 15.693145286143773,
+    "gpu_batches_per_second": 1.0266176407815741,
+    "producer_headroom": 15.286261079827566,
     "sustainable": true,
-    "buffer_seconds": 60.2356415840348,
+    "buffer_seconds": 60.39249428131665,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368168.97172300913,
-    "upload_bytes_per_second": 4022033.258396909,
+    "generation_bytes_per_second": 314210.1598645352,
+    "upload_bytes_per_second": 4060771.463616158,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.00049079673487,
+    "estimated_full_shards_per_hour": 67.42218586875956,
     "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": 1.248541894000482,
-      "bytes": 6395911,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.171668709000187,
       "bytes": 7290583,
@@ -562,6 +556,12 @@
       "bytes": 7163170,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.270394237999426,
+      "bytes": 10514630,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -114,3 +114,4 @@
 {"step": 326400, "loss": 0.6128942668437958, "lr": 0.0003, "grad_norm": 0.15507329924319446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47776104880, "tokens_per_second": 150378.43704593997, "data_wait_s": 1.3608001609100029e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 326500, "loss": 0.6067680865526199, "lr": 0.0003, "grad_norm": 0.15870526442850594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47790775427, "tokens_per_second": 149634.81456336725, "data_wait_s": 1.5564999557682313e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 326600, "loss": 0.5881037451326847, "lr": 0.0003, "grad_norm": 0.16212210363493257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47805508423, "tokens_per_second": 107496.20325382781, "data_wait_s": 1.3380000382312573e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 326700, "loss": 0.5679374486207962, "lr": 0.0003, "grad_norm": 0.14957423102044473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47819987145, "tokens_per_second": 147843.17373110467, "data_wait_s": 1.569199957884848e-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:ee08c2b7aa9cd1c233ae5b0ef594398b978c7962325245493878bb0ecb819ac1
-size 247419
+oid sha256:6e83d2d9a033243bdd26bc22a322f2359b6eca41343756fb014f95a7c95347cf
+size 249537