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": 75475.06458312603,
-      "required_pages_per_second_upper_estimate": 0.006541987308263035
+      "required_token_positions_per_second": 75568.80782593261,
+      "required_pages_per_second_upper_estimate": 0.006550112734959433
     },
     {
       "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": 9434.383072890754,
-      "required_pages_per_second_upper_estimate": 0.0037635164643732067
+      "required_token_positions_per_second": 9446.100978241577,
+      "required_pages_per_second_upper_estimate": 0.0037681909120159474
     },
     {
       "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": 18868.76614578151,
-      "required_pages_per_second_upper_estimate": 0.00436495696888429
+      "required_token_positions_per_second": 18892.201956483153,
+      "required_pages_per_second_upper_estimate": 0.004370378431233935
     },
     {
       "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": 28303.149218672264,
-      "required_pages_per_second_upper_estimate": 0.0017235132922695001
+      "required_token_positions_per_second": 28338.30293472473,
+      "required_pages_per_second_upper_estimate": 0.0017256539691398011
     },
     {
       "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": 28303.149218672264,
-      "required_pages_per_second_upper_estimate": 0.0021788547749849317
+      "required_token_positions_per_second": 28338.30293472473,
+      "required_pages_per_second_upper_estimate": 0.0021815610053583646
     },
     {
       "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": 28303.149218672264,
-      "required_pages_per_second_upper_estimate": 0.3514303887489261
+      "required_token_positions_per_second": 28338.30293472473,
+      "required_pages_per_second_upper_estimate": 0.35186688025037843
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.18298452964069156,
-    "producer_p95_seconds": 0.3307031780022953,
-    "gpu_update_seconds": 0.976852426787898,
-    "producer_batches_per_second": 5.464942866829235,
-    "gpu_batches_per_second": 1.0236960799577641,
-    "producer_headroom": 5.33844270171935,
+    "producer_batch_seconds": 0.0742416452733039,
+    "producer_p95_seconds": 0.269461350999336,
+    "gpu_update_seconds": 0.9756406395854123,
+    "producer_batches_per_second": 13.4695290805952,
+    "gpu_batches_per_second": 1.024967554062671,
+    "producer_headroom": 13.14141996710621,
     "sustainable": true,
-    "buffer_seconds": 60.56485046084968,
+    "buffer_seconds": 60.48971965429556,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326064.18071030086,
+    "generation_bytes_per_second": 325862.137523614,
     "upload_bytes_per_second": 5732622.116984269,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.96578279477853,
+    "estimated_full_shards_per_hour": 69.9224290302402,
     "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": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 20900, "loss": 0.6284867636859417, "lr": 0.0003, "grad_norm": 0.1758226288225565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3058158501, "tokens_per_second": 147687.78176371922, "data_wait_s": 8.006001735338941e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 21000, "loss": 0.6812669988721609, "lr": 0.0003, "grad_norm": 0.13401157085151533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3072774075, "tokens_per_second": 149026.2730084698, "data_wait_s": 1.1579999409150332e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 21100, "loss": 0.6893913336098194, "lr": 0.0003, "grad_norm": 0.1397953381680224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3087368786, "tokens_per_second": 147324.48810078556, "data_wait_s": 1.7758997273631394e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 21200, "loss": 0.6821452751755714, "lr": 0.0003, "grad_norm": 0.14234147294255003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3102061033, "tokens_per_second": 149341.20047718784, "data_wait_s": 1.0720003047026694e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:baabc1f6e96edb26e165ff9e0bd705596a45a3e81f5c26df8060307fcc26f13e
-size 444421
+oid sha256:a1c6ded53593b6b60f56ac5755dacb7dcc2446f21313d5615c5a81df25d7f631
+size 446539