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": 75385.73815857034,
-      "required_pages_per_second_upper_estimate": 0.0065342447201783005
+      "required_token_positions_per_second": 75940.38952669532,
+      "required_pages_per_second_upper_estimate": 0.0065823204950163335
     },
     {
       "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": 9423.217269821293,
-      "required_pages_per_second_upper_estimate": 0.0037590622585851655
+      "required_token_positions_per_second": 9492.548690836915,
+      "required_pages_per_second_upper_estimate": 0.003786719599025417
     },
     {
       "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": 18846.434539642585,
-      "required_pages_per_second_upper_estimate": 0.004359790944826895
+      "required_token_positions_per_second": 18985.09738167383,
+      "required_pages_per_second_upper_estimate": 0.0043918681529481534
     },
     {
       "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": 28269.651809463878,
-      "required_pages_per_second_upper_estimate": 0.001721473475796033
+      "required_token_positions_per_second": 28477.646072510746,
+      "required_pages_per_second_upper_estimate": 0.0017341392351540265
     },
     {
       "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": 28269.651809463878,
-      "required_pages_per_second_upper_estimate": 0.0021762760516973066
+      "required_token_positions_per_second": 28477.646072510746,
+      "required_pages_per_second_upper_estimate": 0.002192288025831625
     },
     {
       "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": 28269.651809463878,
-      "required_pages_per_second_upper_estimate": 0.35101446303517486
+      "required_token_positions_per_second": 28477.646072510746,
+      "required_pages_per_second_upper_estimate": 0.3535970556701981
     }
   ],
   "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": 0.08271581891392543,
-    "producer_p95_seconds": 0.23171673600154463,
-    "gpu_update_seconds": 0.9780099233745863,
-    "producer_batches_per_second": 12.08958592358985,
-    "gpu_batches_per_second": 1.0224845127844284,
-    "producer_headroom": 11.823735002760586,
+    "producer_batch_seconds": 0.062354068367056925,
+    "producer_p95_seconds": 0.21334832000138704,
+    "gpu_update_seconds": 0.9708667608833162,
+    "producer_batches_per_second": 16.03744593076661,
+    "gpu_batches_per_second": 1.030007453432825,
+    "producer_headroom": 15.570223183644698,
     "sustainable": true,
-    "buffer_seconds": 60.63661524922435,
+    "buffer_seconds": 60.19373917476561,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308519.0174445241,
-    "upload_bytes_per_second": 5255244.493300824,
+    "generation_bytes_per_second": 326657.11066888104,
+    "upload_bytes_per_second": 5211679.564971892,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.20099918844025,
+    "estimated_full_shards_per_hour": 70.09301176118683,
     "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": 1.1567702649990679,
-      "bytes": 10579644,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.6480453410003975,
       "bytes": 7697533,
@@ -562,6 +556,12 @@
       "bytes": 2657105,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0683662000010372,
+      "bytes": 6645182,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 212200, "loss": 0.6074743866920471, "lr": 0.0003, "grad_norm": 0.17296962229424495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31058690529, "tokens_per_second": 149128.21955117432, "data_wait_s": 1.2442997103789821e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 212300, "loss": 0.5837424118071795, "lr": 0.0003, "grad_norm": 0.14022896598437734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31073379662, "tokens_per_second": 148958.18816050663, "data_wait_s": 1.9877999875461683e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 212400, "loss": 0.6165460869669914, "lr": 0.0003, "grad_norm": 0.15743356383032697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31088295034, "tokens_per_second": 152686.41461824824, "data_wait_s": 9.823001164477319e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 212500, "loss": 0.6084369644522667, "lr": 0.0003, "grad_norm": 0.14856226591052657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31103080427, "tokens_per_second": 150146.54282454058, "data_wait_s": 1.4111003110883757e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3b6f20ddec4f5ef7eef8e52a50f40203dc675b4a41c4dc1024560b7ba0d11bd
-size 371046
+oid sha256:79f43c8a3b40b0344418ef17513aa668f7f5b1c929c795a75335bb31acf361e2
+size 373164