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": 76580.32436292726,
-      "required_pages_per_second_upper_estimate": 0.005298004530260849
+      "required_token_positions_per_second": 76757.59998872952,
+      "required_pages_per_second_upper_estimate": 0.005310268869389972
     },
     {
       "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": 9572.540545365908,
-      "required_pages_per_second_upper_estimate": 0.00400782282896703
+      "required_token_positions_per_second": 9594.69999859119,
+      "required_pages_per_second_upper_estimate": 0.004017100529290452
     },
     {
       "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": 19145.081090731815,
-      "required_pages_per_second_upper_estimate": 0.00398741550145871
+      "required_token_positions_per_second": 19189.39999718238,
+      "required_pages_per_second_upper_estimate": 0.0039966459609042035
     },
     {
       "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": 28717.621636097723,
-      "required_pages_per_second_upper_estimate": 0.0017215775678147752
+      "required_token_positions_per_second": 28784.09999577357,
+      "required_pages_per_second_upper_estimate": 0.0017255628439707645
     },
     {
       "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": 28717.621636097723,
-      "required_pages_per_second_upper_estimate": 0.002232668766738451
+      "required_token_positions_per_second": 28784.09999577357,
+      "required_pages_per_second_upper_estimate": 0.0022378371667958466
     },
     {
       "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": 28717.621636097723,
-      "required_pages_per_second_upper_estimate": 0.3683588157681113
+      "required_token_positions_per_second": 28784.09999577357,
+      "required_pages_per_second_upper_estimate": 0.36921152878713165
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05129019232811771,
-    "producer_p95_seconds": 0.1705000200008726,
-    "gpu_update_seconds": 0.9627538223864186,
-    "producer_batches_per_second": 19.496904858588174,
-    "gpu_batches_per_second": 1.0386871251482106,
-    "producer_headroom": 18.770719677310097,
+    "producer_batch_seconds": 0.06916123080471692,
+    "producer_p95_seconds": 0.1812615029994049,
+    "gpu_update_seconds": 0.9605302929068344,
+    "producer_batches_per_second": 14.458967666778397,
+    "gpu_batches_per_second": 1.0410915797082454,
+    "producer_headroom": 13.888276448101102,
     "sustainable": true,
-    "buffer_seconds": 60.65349081034437,
+    "buffer_seconds": 60.51340845313057,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395493.58686607966,
-    "upload_bytes_per_second": 2590953.5050091404,
+    "generation_bytes_per_second": 396149.02483672614,
+    "upload_bytes_per_second": 2625602.23827878,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.86371712195198,
+    "estimated_full_shards_per_hour": 85.00435885263765,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 7499226,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2592496409997693,
+      "bytes": 5842307,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 651700, "loss": 0.5561076402664185, "lr": 0.0001, "grad_norm": 0.1972366777198106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95391310126, "tokens_per_second": 151506.7058284033, "data_wait_s": 9.306000720243901e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 651800, "loss": 0.5157055854797363, "lr": 0.0001, "grad_norm": 0.17730778030609043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95405935164, "tokens_per_second": 152620.61596950027, "data_wait_s": 9.216000762535259e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 651900, "loss": 0.5858619809150696, "lr": 0.0001, "grad_norm": 0.17948010360868635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95420643561, "tokens_per_second": 153603.0674927566, "data_wait_s": 9.302000762545504e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 652000, "loss": 0.5466394424438477, "lr": 0.0001, "grad_norm": 0.19738119206916308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95435357325, "tokens_per_second": 153580.7515054078, "data_wait_s": 1.070100006472785e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e30d856525b68f66c77ad22ab10cdc29005d4b33b65ff2434451807ba8368d4
-size 147090
+oid sha256:f150d818195fe6652d743abc7cdb4b8b55582950890895eaf23437f451f2be36
+size 149208