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": 75593.13962539104,
-      "required_pages_per_second_upper_estimate": 0.006552221753668107
+      "required_token_positions_per_second": 75495.39727671558,
+      "required_pages_per_second_upper_estimate": 0.006543749694610636
     },
     {
       "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": 9449.14245317388,
-      "required_pages_per_second_upper_estimate": 0.0037694042018405456
+      "required_token_positions_per_second": 9436.924659589447,
+      "required_pages_per_second_upper_estimate": 0.0037645303413074227
     },
     {
       "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": 18898.28490634776,
-      "required_pages_per_second_upper_estimate": 0.00437178561462885
+      "required_token_positions_per_second": 18873.849319178895,
+      "required_pages_per_second_upper_estimate": 0.00436613287158898
     },
     {
       "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": 28347.42735952164,
-      "required_pages_per_second_upper_estimate": 0.001726209598737776
+      "required_token_positions_per_second": 28310.773978768342,
+      "required_pages_per_second_upper_estimate": 0.0017239776001553267
     },
     {
       "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": 28347.42735952164,
-      "required_pages_per_second_upper_estimate": 0.0021822634288372555
+      "required_token_positions_per_second": 28310.773978768342,
+      "required_pages_per_second_upper_estimate": 0.0021794417501238145
     },
     {
       "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": 28347.42735952164,
-      "required_pages_per_second_upper_estimate": 0.3519801750688707
+      "required_token_positions_per_second": 28310.773978768342,
+      "required_pages_per_second_upper_estimate": 0.35152506275088896
     }
   ],
   "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": 0.49862605985157415,
-    "producer_p95_seconds": 0.3369590160000371,
-    "gpu_update_seconds": 0.9753266019293032,
-    "producer_batches_per_second": 2.0055109038979424,
-    "gpu_batches_per_second": 1.0252975752141797,
-    "producer_headroom": 1.9560281350309454,
+    "producer_batch_seconds": 0.10599076562513687,
+    "producer_p95_seconds": 0.3589136119990144,
+    "gpu_update_seconds": 0.9765893373573824,
+    "producer_batches_per_second": 9.434784191830001,
+    "gpu_batches_per_second": 1.02397185976448,
+    "producer_headroom": 9.213909642009167,
     "sustainable": true,
-    "buffer_seconds": 60.4702493196168,
+    "buffer_seconds": 60.54853891615771,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 236313.03798811714,
-    "upload_bytes_per_second": 3763948.2732370384,
+    "generation_bytes_per_second": 234829.2531983327,
+    "upload_bytes_per_second": 3753817.571020143,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.707276866270405,
+    "estimated_full_shards_per_hour": 50.388891191124785,
     "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": 1.5801954900016426,
-      "bytes": 8475194,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.4765983690012945,
       "bytes": 10921563,
@@ -562,6 +556,12 @@
       "bytes": 13118636,
       "pages": 23,
       "shards": 4
+    },
+    {
+      "seconds": 1.014325338001072,
+      "bytes": 5232610,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -302,3 +302,4 @@
 {"step": 225200, "loss": 0.6066942475736141, "lr": 0.0003, "grad_norm": 0.14503524494823936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32962651332, "tokens_per_second": 117636.08793781688, "data_wait_s": 1.7193000530824065e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225300, "loss": 0.6036153789609671, "lr": 0.0003, "grad_norm": 0.14570894164073353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32977525139, "tokens_per_second": 127764.831629876, "data_wait_s": 1.0690000635804608e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 225400, "loss": 0.5949903167784214, "lr": 0.0003, "grad_norm": 0.1468773885388837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32992228475, "tokens_per_second": 149077.31021027215, "data_wait_s": 1.2499000149546191e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 225500, "loss": 0.579850235953927, "lr": 0.0003, "grad_norm": 0.155292803204275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33007004158, "tokens_per_second": 150117.14766228411, "data_wait_s": 2.019899693550542e-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 sha256:5e99108b87fc7fd76cd00560400e3563ba24c52e30a8e533af198ee7e45cd6c1
-size 648735
+oid sha256:0ff5b4e865feab96760da84c8c929b6aff72bcafb479d8acf587bf1ab4b80239
+size 650853