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": 75823.54917415562,
-      "required_pages_per_second_upper_estimate": 0.005245649066741265
+      "required_token_positions_per_second": 75905.59197278159,
+      "required_pages_per_second_upper_estimate": 0.00525132497791573
     },
     {
       "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": 9477.943646769452,
-      "required_pages_per_second_upper_estimate": 0.0039682170829325675
+      "required_token_positions_per_second": 9488.198996597699,
+      "required_pages_per_second_upper_estimate": 0.003972510783749598
     },
     {
       "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": 18955.887293538904,
-      "required_pages_per_second_upper_estimate": 0.0039480114228793795
+      "required_token_positions_per_second": 18976.397993195398,
+      "required_pages_per_second_upper_estimate": 0.0039522832607143035
     },
     {
       "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": 28433.83094030836,
-      "required_pages_per_second_upper_estimate": 0.0017045647489255052
+      "required_token_positions_per_second": 28464.596989793095,
+      "required_pages_per_second_upper_estimate": 0.0017064091266150775
     },
     {
       "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": 28433.83094030836,
-      "required_pages_per_second_upper_estimate": 0.002210605288404178
+      "required_token_positions_per_second": 28464.596989793095,
+      "required_pages_per_second_upper_estimate": 0.0022129972134260663
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28433.83094030836,
-      "required_pages_per_second_upper_estimate": 0.364718653433234
+      "required_token_positions_per_second": 28464.596989793095,
+      "required_pages_per_second_upper_estimate": 0.36511328728201403
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08336772118752833,
-    "producer_p95_seconds": 0.23506004999944707,
-    "gpu_update_seconds": 0.9723628187155622,
-    "producer_batches_per_second": 11.995050191555412,
-    "gpu_batches_per_second": 1.0284227047275882,
-    "producer_headroom": 11.663540814895466,
+    "producer_batch_seconds": 0.46736546910159404,
+    "producer_p95_seconds": 0.2166294660000858,
+    "gpu_update_seconds": 0.9713118372943796,
+    "producer_batches_per_second": 2.139653153927433,
+    "gpu_batches_per_second": 1.02953548140166,
+    "producer_headroom": 2.078270436113969,
     "sustainable": true,
-    "buffer_seconds": 60.28649476036485,
+    "buffer_seconds": 60.22133391225154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311305.1667836292,
-    "upload_bytes_per_second": 4818350.120477331,
+    "generation_bytes_per_second": 295003.63092794194,
+    "upload_bytes_per_second": 4825120.841091904,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.79884197837502,
+    "estimated_full_shards_per_hour": 63.300911864077506,
     "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": 1.466896263000308,
-      "bytes": 7685001,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0174859720000313,
       "bytes": 4930803,
@@ -562,6 +556,12 @@
       "bytes": 7665740,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1344610849992023,
+      "bytes": 6680190,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 735400, "loss": 0.5374253392219543, "lr": 5e-05, "grad_norm": 0.23699762739642305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107642136451, "tokens_per_second": 150306.35628847012, "data_wait_s": 9.063000106834807e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 735500, "loss": 0.5342980623245239, "lr": 5e-05, "grad_norm": 0.22605024981864008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107656763987, "tokens_per_second": 150737.127845136, "data_wait_s": 1.4728999303770252e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 735600, "loss": 0.5117169618606567, "lr": 5e-05, "grad_norm": 0.2415996822182658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107671466857, "tokens_per_second": 130425.02266456, "data_wait_s": 9.953000699169934e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 735700, "loss": 0.5571478009223938, "lr": 5e-05, "grad_norm": 0.23093229228948703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107686059365, "tokens_per_second": 151357.19492949295, "data_wait_s": 1.6595999113633297e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba57503acc146c3ed1959625c792b256850437fa7f00575c9a2ec8b1065bbf0e
-size 332922
+oid sha256:08b6d03577b05f3dfafe16ef43ac315c7675e7498391535f6691375aa8823262
+size 335040