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": 75844.36182502533,
-      "required_pages_per_second_upper_estimate": 0.006573997057215564
+      "required_token_positions_per_second": 75358.37297923162,
+      "required_pages_per_second_upper_estimate": 0.006531872775789633
     },
     {
       "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": 9480.545228128167,
-      "required_pages_per_second_upper_estimate": 0.003781931238283137
+      "required_token_positions_per_second": 9419.796622403952,
+      "required_pages_per_second_upper_estimate": 0.0037576977111871517
     },
     {
       "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": 18961.090456256334,
-      "required_pages_per_second_upper_estimate": 0.004386314573260272
+      "required_token_positions_per_second": 18839.593244807904,
+      "required_pages_per_second_upper_estimate": 0.004358208331669568
     },
     {
       "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": 28441.6356843845,
-      "required_pages_per_second_upper_estimate": 0.001731946391449043
+      "required_token_positions_per_second": 28259.389867211856,
+      "required_pages_per_second_upper_estimate": 0.001720848577353135
     },
     {
       "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": 28441.6356843845,
-      "required_pages_per_second_upper_estimate": 0.0021895158464705325
+      "required_token_positions_per_second": 28259.389867211856,
+      "required_pages_per_second_upper_estimate": 0.002175486058975872
     },
     {
       "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": 28441.6356843845,
-      "required_pages_per_second_upper_estimate": 0.35314992716868643
+      "required_token_positions_per_second": 28259.389867211856,
+      "required_pages_per_second_upper_estimate": 0.3508870440569161
     }
   ],
   "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": 0.4554932262265936,
-    "producer_p95_seconds": 0.2936086739973689,
-    "gpu_update_seconds": 0.9720959900762587,
-    "producer_batches_per_second": 2.1954223299525673,
-    "gpu_batches_per_second": 1.0287049943715458,
-    "producer_headroom": 2.1341612434707677,
+    "producer_batch_seconds": 0.12220810110162006,
+    "producer_p95_seconds": 0.2566114019973611,
+    "gpu_update_seconds": 0.9783650719253063,
+    "producer_batches_per_second": 8.1827635892032,
+    "gpu_batches_per_second": 1.022113348785151,
+    "producer_headroom": 8.005730087498566,
     "sustainable": true,
-    "buffer_seconds": 60.26995138472804,
+    "buffer_seconds": 60.65863445936899,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297733.22909116064,
-    "upload_bytes_per_second": 5235835.227324252,
+    "generation_bytes_per_second": 276071.79502444115,
+    "upload_bytes_per_second": 5290750.947078448,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.88662008811106,
+    "estimated_full_shards_per_hour": 59.238580589770564,
     "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": 1.2853153289997863,
-      "bytes": 6668992,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.2017098229989642,
       "bytes": 7521074,
@@ -562,6 +556,12 @@
       "bytes": 6719573,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.3891484439991473,
+      "bytes": 11561860,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -174,3 +174,4 @@
 {"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}
 {"step": 212600, "loss": 0.6178873274475336, "lr": 0.0003, "grad_norm": 0.1360110814119623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31117767975, "tokens_per_second": 148711.80802560845, "data_wait_s": 1.4929999451851472e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 212700, "loss": 0.5949128679931164, "lr": 0.0003, "grad_norm": 0.13667872401146625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31132442379, "tokens_per_second": 149344.60654691805, "data_wait_s": 1.8855000234907493e-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:bab3b18034a17157b3836cc85595d8b13366d0049d96fed798082b7fccb3a704
-size 375282
+oid sha256:0c7f9142d4a81840a29de5cb15926baef5c61f3e487c01734d763d84fd5e263a
+size 377400