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": 75274.53491077588,
-      "required_pages_per_second_upper_estimate": 0.006524605904501532
+      "required_token_positions_per_second": 75598.5316342505,
+      "required_pages_per_second_upper_estimate": 0.0065526891193301235
     },
     {
       "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": 9409.316863846985,
-      "required_pages_per_second_upper_estimate": 0.003753517178812424
+      "required_token_positions_per_second": 9449.816454281312,
+      "required_pages_per_second_upper_estimate": 0.003769673070959515
     },
     {
       "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": 18818.63372769397,
-      "required_pages_per_second_upper_estimate": 0.0043533597162601625
+      "required_token_positions_per_second": 18899.632908562624,
+      "required_pages_per_second_upper_estimate": 0.00437209745121723
     },
     {
       "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": 28227.950591540954,
-      "required_pages_per_second_upper_estimate": 0.0017189340904139087
+      "required_token_positions_per_second": 28349.449362843934,
+      "required_pages_per_second_upper_estimate": 0.0017263327281314721
     },
     {
       "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": 28227.950591540954,
-      "required_pages_per_second_upper_estimate": 0.0021730657765052404
+      "required_token_positions_per_second": 28349.449362843934,
+      "required_pages_per_second_upper_estimate": 0.002182419088250269
     },
     {
       "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": 28227.950591540954,
-      "required_pages_per_second_upper_estimate": 0.35049667347357055
+      "required_token_positions_per_second": 28349.449362843934,
+      "required_pages_per_second_upper_estimate": 0.3520052815829238
     }
   ],
   "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": 0.10111107481219506,
-    "producer_p95_seconds": 0.3328930819989182,
-    "gpu_update_seconds": 0.9794547397388903,
-    "producer_batches_per_second": 9.890113440663272,
-    "gpu_batches_per_second": 1.020976222205619,
-    "producer_headroom": 9.686918486012946,
+    "producer_batch_seconds": 0.263509006851649,
+    "producer_p95_seconds": 0.32680167599755805,
+    "gpu_update_seconds": 0.9752570374871801,
+    "producer_batches_per_second": 3.794936696653343,
+    "gpu_batches_per_second": 1.0253707090148994,
+    "producer_headroom": 3.7010387202295245,
     "sustainable": true,
-    "buffer_seconds": 60.7261938638112,
+    "buffer_seconds": 60.46593632420517,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349545.3998285086,
-    "upload_bytes_per_second": 4425108.101236293,
+    "generation_bytes_per_second": 372935.845468488,
+    "upload_bytes_per_second": 4393328.842840834,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.00430580274052,
+    "estimated_full_shards_per_hour": 80.02335093537312,
     "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": 1.3386981250005192,
-      "bytes": 10151331,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0260517099995923,
       "bytes": 2406820,
@@ -562,6 +556,12 @@
       "bytes": 2478260,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1128817630014964,
+      "bytes": 6162674,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -276,3 +276,4 @@
 {"step": 264700, "loss": 0.5824983902275562, "lr": 0.0003, "grad_norm": 0.16188175198946875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38744856079, "tokens_per_second": 148505.21341425242, "data_wait_s": 1.5558001905446872e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 264800, "loss": 0.5940949842333794, "lr": 0.0003, "grad_norm": 0.15285461701734704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38759615411, "tokens_per_second": 150145.6373441198, "data_wait_s": 1.5038000128697604e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 264900, "loss": 0.6083609499037266, "lr": 0.0003, "grad_norm": 0.1712524332294461, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38774309209, "tokens_per_second": 148624.64394045892, "data_wait_s": 1.2454001989681274e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 265000, "loss": 0.5934735797345638, "lr": 0.0003, "grad_norm": 0.15487432055418565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38789095721, "tokens_per_second": 150745.35371688954, "data_wait_s": 1.3623000995721668e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5336505d168c88d6797fbaff21c90a203b98abdec74f7c13ee986e244d0e4029
-size 592884
+oid sha256:3b6881d419716d42c466b3c038d823f46c8d38f74b489b60d0f8335132bdcc1a
+size 595002