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": 75749.173595106,
-      "required_pages_per_second_upper_estimate": 0.006565746382698519
+      "required_token_positions_per_second": 75641.68976087315,
+      "required_pages_per_second_upper_estimate": 0.006556429956362503
     },
     {
       "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": 9468.64669938825,
-      "required_pages_per_second_upper_estimate": 0.0037771847372699255
+      "required_token_positions_per_second": 9455.211220109144,
+      "required_pages_per_second_upper_estimate": 0.003771825123707174
     },
     {
       "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": 18937.2933987765,
-      "required_pages_per_second_upper_estimate": 0.00438080954282622
+      "required_token_positions_per_second": 18910.422440218288,
+      "required_pages_per_second_upper_estimate": 0.0043745934194764965
     },
     {
       "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": 28405.94009816475,
-      "required_pages_per_second_upper_estimate": 0.0017297727175285273
+      "required_token_positions_per_second": 28365.633660327432,
+      "required_pages_per_second_upper_estimate": 0.0017273182669357706
     },
     {
       "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": 28405.94009816475,
-      "required_pages_per_second_upper_estimate": 0.002186767901431629
+      "required_token_positions_per_second": 28365.633660327432,
+      "required_pages_per_second_upper_estimate": 0.0021836650002715514
     },
     {
       "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": 28405.94009816475,
-      "required_pages_per_second_upper_estimate": 0.3527067074532792
+      "required_token_positions_per_second": 28365.633660327432,
+      "required_pages_per_second_upper_estimate": 0.3522062363923095
     }
   ],
   "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": 0.08826508694525614,
-    "producer_p95_seconds": 0.24785237899777712,
-    "gpu_update_seconds": 0.9733175492328199,
-    "producer_batches_per_second": 11.329507901807437,
-    "gpu_batches_per_second": 1.027413921374593,
-    "producer_headroom": 11.027208865001082,
+    "producer_batch_seconds": 0.09823402832034844,
+    "producer_p95_seconds": 0.2323412969999481,
+    "gpu_update_seconds": 0.9747005947788459,
+    "producer_batches_per_second": 10.179771888605911,
+    "gpu_batches_per_second": 1.0259560785708708,
+    "producer_headroom": 9.922229714537156,
     "sustainable": true,
-    "buffer_seconds": 60.34568805243483,
+    "buffer_seconds": 60.431436876288444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 194665.8032566004,
+    "generation_bytes_per_second": 191667.16212684594,
     "upload_bytes_per_second": 6237772.532715363,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 41.77074979089269,
+    "estimated_full_shards_per_hour": 41.12731120924028,
     "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

@@ -166,3 +166,4 @@
 {"step": 16600, "loss": 0.6558071514591575, "lr": 0.0003, "grad_norm": 0.1453104804673411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2428926232, "tokens_per_second": 149604.24118273135, "data_wait_s": 1.0735999239841476e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 16700, "loss": 0.6939180009067059, "lr": 0.0003, "grad_norm": 0.13559673900900088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2443612035, "tokens_per_second": 149759.9458962013, "data_wait_s": 1.0638003004714847e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 16800, "loss": 0.685417115688324, "lr": 0.0003, "grad_norm": 0.14467339365779758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2458324645, "tokens_per_second": 120802.5654527288, "data_wait_s": 1.2920998415211216e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 16900, "loss": 0.6956988908350468, "lr": 0.0003, "grad_norm": 0.18371533973157148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2473089397, "tokens_per_second": 150360.3606912104, "data_wait_s": 1.3584001862909645e-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:d4e7755edbb5ee4bc2005fa554b4a61d538308130453720a6ab210d6f8f93fd7
-size 352564
+oid sha256:9532e65fa5704f458edb2e3d40c262b9d0658d287b07d3e074fefa5f66d22b9e
+size 354682