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": 76040.78991773669,
-      "required_pages_per_second_upper_estimate": 0.006591022946449335
+      "required_token_positions_per_second": 75536.28861898786,
+      "required_pages_per_second_upper_estimate": 0.006547294052520638
     },
     {
       "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": 9505.098739717087,
-      "required_pages_per_second_upper_estimate": 0.0037917260011636695
+      "required_token_positions_per_second": 9442.036077373483,
+      "required_pages_per_second_upper_estimate": 0.0037665693622839806
     },
     {
       "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": 19010.197479434173,
-      "required_pages_per_second_upper_estimate": 0.004397674618818376
+      "required_token_positions_per_second": 18884.072154746966,
+      "required_pages_per_second_upper_estimate": 0.0043684977446818915
     },
     {
       "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": 28515.29621915126,
-      "required_pages_per_second_upper_estimate": 0.0017364319315493805
+      "required_token_positions_per_second": 28326.10823212045,
+      "required_pages_per_second_upper_estimate": 0.0017249113757318605
     },
     {
       "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": 28515.29621915126,
-      "required_pages_per_second_upper_estimate": 0.0021951864383422883
+      "required_token_positions_per_second": 28326.10823212045,
+      "required_pages_per_second_upper_estimate": 0.0021806222233947898
     },
     {
       "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": 28515.29621915126,
-      "required_pages_per_second_upper_estimate": 0.3540645444845383
+      "required_token_positions_per_second": 28326.10823212045,
+      "required_pages_per_second_upper_estimate": 0.3517154628570775
     }
   ],
   "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": 0.11734972100784091,
-    "producer_p95_seconds": 0.21089364199906413,
-    "gpu_update_seconds": 0.9695848777973145,
-    "producer_batches_per_second": 8.521537089408021,
-    "gpu_batches_per_second": 1.031369220889441,
-    "producer_headroom": 8.26235349747896,
+    "producer_batch_seconds": 0.2476910956952878,
+    "producer_p95_seconds": 0.22579208100069081,
+    "gpu_update_seconds": 0.976060663661819,
+    "producer_batches_per_second": 4.037286835818315,
+    "gpu_batches_per_second": 1.0245264840900046,
+    "producer_headroom": 3.94063686836195,
     "sustainable": true,
-    "buffer_seconds": 60.1142624234335,
+    "buffer_seconds": 60.51576114703278,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430838.5445086987,
-    "upload_bytes_per_second": 5993876.044067141,
+    "generation_bytes_per_second": 415704.6575034059,
+    "upload_bytes_per_second": 6149821.470645785,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.44792224355432,
+    "estimated_full_shards_per_hour": 89.20054239107735,
     "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": 0.999679897000533,
-      "bytes": 1109612,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 0.9920720150003035,
       "bytes": 6122911,
@@ -562,6 +556,12 @@
       "bytes": 7770470,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.5646253450013319,
+      "bytes": 15407220,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -116,3 +116,4 @@
 {"step": 51600, "loss": 0.6494944877922535, "lr": 0.0003, "grad_norm": 0.13493877725330178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7552234544, "tokens_per_second": 151325.46088628046, "data_wait_s": 1.1256999641773291e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 51700, "loss": 0.6375000514090061, "lr": 0.0003, "grad_norm": 0.13611204968969093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7567016266, "tokens_per_second": 150654.87974641632, "data_wait_s": 1.2184998922748491e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 51800, "loss": 0.5903803128749132, "lr": 0.0003, "grad_norm": 0.1342751621928546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7581571065, "tokens_per_second": 147936.7649757513, "data_wait_s": 1.055899883795064e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 51900, "loss": 0.6394411474466324, "lr": 0.0003, "grad_norm": 0.1356565594248017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7596081523, "tokens_per_second": 148360.54984097305, "data_wait_s": 7.036000170046464e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc4b92a425df486af251188832591a37299621b1b411448a89ef0ec740154a8a
-size 251655
+oid sha256:27b4818f57d7c22354d531aa2fdc3a9efec1ac6824f7c74600fcddb8ff589bab
+size 253773