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": 75593.83277893273,
-      "required_pages_per_second_upper_estimate": 0.0065522818344603195
+      "required_token_positions_per_second": 75440.2719060439,
+      "required_pages_per_second_upper_estimate": 0.00653897156719479
     },
     {
       "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": 9449.229097366591,
-      "required_pages_per_second_upper_estimate": 0.0037694387655044645
+      "required_token_positions_per_second": 9430.033988255487,
+      "required_pages_per_second_upper_estimate": 0.0037617815494875887
     },
     {
       "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": 18898.458194733183,
-      "required_pages_per_second_upper_estimate": 0.004371825701847047
+      "required_token_positions_per_second": 18860.067976510974,
+      "required_pages_per_second_upper_estimate": 0.004362944800506103
     },
     {
       "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": 28347.687292099774,
-      "required_pages_per_second_upper_estimate": 0.0017262254272680235
+      "required_token_positions_per_second": 28290.101964766458,
+      "required_pages_per_second_upper_estimate": 0.0017227187829602876
     },
     {
       "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": 28347.687292099774,
-      "required_pages_per_second_upper_estimate": 0.0021822834391666593
+      "required_token_positions_per_second": 28290.101964766458,
+      "required_pages_per_second_upper_estimate": 0.0021778503612621524
     },
     {
       "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": 28347.687292099774,
-      "required_pages_per_second_upper_estimate": 0.35198340256155275
+      "required_token_positions_per_second": 28290.101964766458,
+      "required_pages_per_second_upper_estimate": 0.35126838552176587
     }
   ],
   "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": 0.11728121791418289,
-    "producer_p95_seconds": 0.43075106700416654,
-    "gpu_update_seconds": 0.9753176587250287,
-    "producer_batches_per_second": 8.526514456319177,
-    "gpu_batches_per_second": 1.0253069767107845,
-    "producer_headroom": 8.31606011662233,
+    "producer_batch_seconds": 0.20856630120351838,
+    "producer_p95_seconds": 0.40217529499932425,
+    "gpu_update_seconds": 0.9773029462542709,
+    "producer_batches_per_second": 4.794638415839781,
+    "gpu_batches_per_second": 1.023224174072861,
+    "producer_headroom": 4.685814250024128,
     "sustainable": true,
-    "buffer_seconds": 60.46969484095178,
+    "buffer_seconds": 60.5927826677648,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493046.2483939648,
-    "upload_bytes_per_second": 5806976.598325157,
+    "generation_bytes_per_second": 484650.8380423174,
+    "upload_bytes_per_second": 5855305.112275578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.79624737610061,
+    "estimated_full_shards_per_hour": 103.99478774978773,
     "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": 1.0406464490006329,
-      "bytes": 6219293,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1162217770033749,
       "bytes": 6495200,
@@ -562,6 +556,12 @@
       "bytes": 2616949,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.138771516001725,
+      "bytes": 10215217,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 155100, "loss": 0.5673714242875576, "lr": 0.0003, "grad_norm": 0.13810040368444318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22698961668, "tokens_per_second": 134032.77837987136, "data_wait_s": 1.4806995750404894e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155200, "loss": 0.6262595262378454, "lr": 0.0003, "grad_norm": 0.14756991405259748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22713619019, "tokens_per_second": 148799.30874215477, "data_wait_s": 1.1732998245861381e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155300, "loss": 0.622527826577425, "lr": 0.0003, "grad_norm": 0.16087515427852186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22728190005, "tokens_per_second": 148964.77667606197, "data_wait_s": 9.969997336156666e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 155400, "loss": 0.619165938347578, "lr": 0.0003, "grad_norm": 0.1476237453690166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22742793235, "tokens_per_second": 148852.73737805066, "data_wait_s": 1.5402001736219972e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:106e4b565ee9f3a764b3623cc4593e1d07ddc70224feea63e37df8fb9daa6df7
-size 859983
+oid sha256:0ee35c15a7fcd35e49d8d0fa0588893f8713cf34ad42b3296d1050880ecf2de8
+size 862101