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": 77015.94124481767,
-      "required_pages_per_second_upper_estimate": 0.005328141516920451
+      "required_token_positions_per_second": 76870.53137460076,
+      "required_pages_per_second_upper_estimate": 0.005318081724701462
     },
     {
       "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": 9626.99265560221,
-      "required_pages_per_second_upper_estimate": 0.004030620790433604
+      "required_token_positions_per_second": 9608.816421825095,
+      "required_pages_per_second_upper_estimate": 0.00402301078091405
     },
     {
       "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": 19253.98531120442,
-      "required_pages_per_second_upper_estimate": 0.004010097378585726
+      "required_token_positions_per_second": 19217.63284365019,
+      "required_pages_per_second_upper_estimate": 0.004002526118273218
     },
     {
       "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": 28880.977966806626,
-      "required_pages_per_second_upper_estimate": 0.0017313705304100994
+      "required_token_positions_per_second": 28826.449265475287,
+      "required_pages_per_second_upper_estimate": 0.0017281016180257917
     },
     {
       "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": 28880.977966806626,
-      "required_pages_per_second_upper_estimate": 0.0022453690029225057
+      "required_token_positions_per_second": 28826.449265475287,
+      "required_pages_per_second_upper_estimate": 0.002241129636240404
     },
     {
       "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": 28880.977966806626,
-      "required_pages_per_second_upper_estimate": 0.3704541753800827
+      "required_token_positions_per_second": 28826.449265475287,
+      "required_pages_per_second_upper_estimate": 0.3697547397477622
     }
   ],
   "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": 0.30618901764859174,
-    "producer_p95_seconds": 0.28526349399908213,
-    "gpu_update_seconds": 0.9573083027789534,
-    "producer_batches_per_second": 3.2659564594432453,
-    "gpu_batches_per_second": 1.0445955572485035,
-    "producer_headroom": 3.126527235139573,
+    "producer_batch_seconds": 0.09273809576569647,
+    "producer_p95_seconds": 0.2675955000013346,
+    "gpu_update_seconds": 0.9591191667547246,
+    "producer_batches_per_second": 10.78305513762659,
+    "gpu_batches_per_second": 1.042623309659841,
+    "producer_headroom": 10.342234858670666,
     "sustainable": true,
-    "buffer_seconds": 60.31042307507406,
+    "buffer_seconds": 60.42450750554765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 607055.3556032173,
-    "upload_bytes_per_second": 5393998.013772451,
+    "generation_bytes_per_second": 610043.1801494232,
+    "upload_bytes_per_second": 5368129.427869143,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.12997389351077,
+    "estimated_full_shards_per_hour": 65.4505326908208,
     "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": 1.2241392690011708,
-      "bytes": 7522866,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.1316358820004098,
       "bytes": 5807643,
@@ -562,6 +556,12 @@
       "bytes": 17178237,
       "pages": 40,
       "shards": 5
+    },
+    {
+      "seconds": 1.2973765460010327,
+      "bytes": 5824332,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -252,3 +252,4 @@
 {"step": 670200, "loss": 0.5137597918510437, "lr": 0.0001, "grad_norm": 0.1884368122650523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98100440735, "tokens_per_second": 150034.85117115034, "data_wait_s": 1.516799966339022e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 670300, "loss": 0.5634480118751526, "lr": 0.0001, "grad_norm": 0.22470139104717005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98115046803, "tokens_per_second": 151468.72484252273, "data_wait_s": 8.707000233698636e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 670400, "loss": 0.4927593171596527, "lr": 0.0001, "grad_norm": 0.1783781125813963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98129679618, "tokens_per_second": 151830.30857302772, "data_wait_s": 1.2854001397499815e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 670500, "loss": 0.5398595333099365, "lr": 0.0001, "grad_norm": 0.18922684276234444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98144384948, "tokens_per_second": 153014.43667492302, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55ad84b5841e70efb3bc54b8a4fd4c68b277103d979382ebf8a30c7f88fb6f23
-size 542052
+oid sha256:3654845a084e12997b417059c02aca377debe7179c02d2b4999ea6a6783e42e8
+size 544170