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": 75261.41216269048,
-      "required_pages_per_second_upper_estimate": 0.006523468457956791
+      "required_token_positions_per_second": 75110.71524930363,
+      "required_pages_per_second_upper_estimate": 0.006510406431442253
     },
     {
       "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": 9407.67652033631,
-      "required_pages_per_second_upper_estimate": 0.003752862821260695
+      "required_token_positions_per_second": 9388.839406162953,
+      "required_pages_per_second_upper_estimate": 0.003745348414776988
     },
     {
       "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": 18815.35304067262,
-      "required_pages_per_second_upper_estimate": 0.0043526007870558926
+      "required_token_positions_per_second": 18777.678812325907,
+      "required_pages_per_second_upper_estimate": 0.004343885517371654
     },
     {
       "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": 28223.02956100893,
-      "required_pages_per_second_upper_estimate": 0.00171863442547316
+      "required_token_positions_per_second": 28166.51821848886,
+      "required_pages_per_second_upper_estimate": 0.001715193181205788
     },
     {
       "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": 28223.02956100893,
-      "required_pages_per_second_upper_estimate": 0.002172686941952601
+      "required_token_positions_per_second": 28166.51821848886,
+      "required_pages_per_second_upper_estimate": 0.002168336542371998
     },
     {
       "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": 28223.02956100893,
-      "required_pages_per_second_upper_estimate": 0.35043557074399256
+      "required_token_positions_per_second": 28166.51821848886,
+      "required_pages_per_second_upper_estimate": 0.3497338890011903
     }
   ],
   "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": 0.10369241064063317,
-    "producer_p95_seconds": 0.3559326480026357,
-    "gpu_update_seconds": 0.9796255196570621,
-    "producer_batches_per_second": 9.643907339233344,
-    "gpu_batches_per_second": 1.0207982335434365,
-    "producer_headroom": 9.44741773872102,
+    "producer_batch_seconds": 0.10303449064059578,
+    "producer_p95_seconds": 0.3434961779930745,
+    "gpu_update_seconds": 0.981590972144065,
+    "producer_batches_per_second": 9.705487878696788,
+    "gpu_batches_per_second": 1.0187542758423345,
+    "producer_headroom": 9.52681928198242,
     "sustainable": true,
-    "buffer_seconds": 60.73678221873785,
+    "buffer_seconds": 60.85864027293203,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315481.3376733986,
-    "upload_bytes_per_second": 5704820.221651031,
+    "generation_bytes_per_second": 301729.1464501961,
+    "upload_bytes_per_second": 5662886.219099597,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.69495103503674,
+    "estimated_full_shards_per_hour": 64.74405093316471,
     "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": 1.1862465560006967,
-      "bytes": 10217288,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.3918344490011805,
       "bytes": 7389964,
@@ -562,6 +556,12 @@
       "bytes": 1128633,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.9270145089976722,
+      "bytes": 5635685,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 306600, "loss": 0.6092037744820118, "lr": 0.0003, "grad_norm": 0.1592246802965326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44877971060, "tokens_per_second": 149919.76741815222, "data_wait_s": 1.1723001080099493e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 306700, "loss": 0.5748039167374372, "lr": 0.0003, "grad_norm": 0.14108368454578674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44892599415, "tokens_per_second": 150016.32332066845, "data_wait_s": 1.2319003872107714e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 306800, "loss": 0.5902784466743469, "lr": 0.0003, "grad_norm": 0.13772920403622205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44907230041, "tokens_per_second": 150036.7816876341, "data_wait_s": 9.659997886046767e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 306900, "loss": 0.5960936769843102, "lr": 0.0003, "grad_norm": 0.15019100022239507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44921851435, "tokens_per_second": 150143.078845326, "data_wait_s": 9.848001354839653e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:56b7f09cb8c4924f3f485257d07a7161e6c6469cbd340f93bbf0b83232f4bdc8
-size 567468
+oid sha256:0c06b13979003552156eacb577a87f5a827aaff10699fa898f435636b844bbed
+size 569586