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": 77303.44125654947,
-      "required_pages_per_second_upper_estimate": 0.0053480314348239915
+      "required_token_positions_per_second": 77159.19832492783,
+      "required_pages_per_second_upper_estimate": 0.005338052374124693
     },
     {
       "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": 9662.930157068684,
-      "required_pages_per_second_upper_estimate": 0.004045667071837249
+      "required_token_positions_per_second": 9644.899790615978,
+      "required_pages_per_second_upper_estimate": 0.004038118133920368
     },
     {
       "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": 19325.860314137368,
-      "required_pages_per_second_upper_estimate": 0.004025067046225367
+      "required_token_positions_per_second": 19289.799581231957,
+      "required_pages_per_second_upper_estimate": 0.004017556546546648
     },
     {
       "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": 28988.79047120605,
-      "required_pages_per_second_upper_estimate": 0.0017378337254286827
+      "required_token_positions_per_second": 28934.699371847935,
+      "required_pages_per_second_upper_estimate": 0.00173459104661449
     },
     {
       "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": 28988.79047120605,
-      "required_pages_per_second_upper_estimate": 0.002253750950922471
+      "required_token_positions_per_second": 28934.699371847935,
+      "required_pages_per_second_upper_estimate": 0.0022495456058689795
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28988.79047120605,
-      "required_pages_per_second_upper_estimate": 0.3718370784264703
+      "required_token_positions_per_second": 28934.699371847935,
+      "required_pages_per_second_upper_estimate": 0.37114325588240193
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.057089339765653335,
-    "producer_p95_seconds": 0.19905558800019207,
-    "gpu_update_seconds": 0.9537479677692027,
-    "producer_batches_per_second": 17.51640506099582,
-    "gpu_batches_per_second": 1.0484950257235985,
-    "producer_headroom": 16.70623572954694,
+    "producer_batch_seconds": 0.09993768920314494,
+    "producer_p95_seconds": 0.1958379120005702,
+    "gpu_update_seconds": 0.9555309230860773,
+    "producer_batches_per_second": 10.00623496474172,
+    "gpu_batches_per_second": 1.0465386057526018,
+    "producer_headroom": 9.561266932475839,
     "sustainable": true,
-    "buffer_seconds": 60.08612196945977,
+    "buffer_seconds": 60.19844815442287,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291965.4504396754,
-    "upload_bytes_per_second": 4884452.227557347,
+    "generation_bytes_per_second": 299906.6017006437,
+    "upload_bytes_per_second": 4884652.560857204,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.64898905651757,
+    "estimated_full_shards_per_hour": 64.35297525658115,
     "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": 1.4142575109999598,
-      "bytes": 6051595,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.509009006000042,
       "bytes": 6599366,
@@ -562,6 +556,12 @@
       "bytes": 5284765,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.747680902999491,
+      "bytes": 7697898,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 851300, "loss": 0.5254843235015869, "lr": 3e-05, "grad_norm": 0.294936959301028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124606475468, "tokens_per_second": 150537.61678752638, "data_wait_s": 8.84500150277745e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 851400, "loss": 0.522243320941925, "lr": 3e-05, "grad_norm": 0.2925135480680224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124621075859, "tokens_per_second": 152220.8878887611, "data_wait_s": 1.1805999747593887e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 851500, "loss": 0.5021769404411316, "lr": 3e-05, "grad_norm": 0.29591102837388067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124635617791, "tokens_per_second": 151674.04648592143, "data_wait_s": 1.0235000445391051e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 851600, "loss": 0.5318317413330078, "lr": 3e-05, "grad_norm": 0.3137795975406802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124650436349, "tokens_per_second": 154114.831035925, "data_wait_s": 1.3015000149607658e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78366a831280d893f7dd01b2c7d93e8e6eb87a87d98cf641caa46ed64cf6b9d8
-size 192351
+oid sha256:6aa7f15a6c9e140a29a4aaff0261f3547ad5a9e4576d15f2054a350cfea7140a
+size 194469