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": 75583.09720847048,
-      "required_pages_per_second_upper_estimate": 0.006551351302421706
+      "required_token_positions_per_second": 76023.60338509848,
+      "required_pages_per_second_upper_estimate": 0.006589533261359123
     },
     {
       "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": 9447.88715105881,
-      "required_pages_per_second_upper_estimate": 0.003768903443058405
+      "required_token_positions_per_second": 9502.95042313731,
+      "required_pages_per_second_upper_estimate": 0.0037908690055598014
     },
     {
       "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": 18895.77430211762,
-      "required_pages_per_second_upper_estimate": 0.004371204830525333
+      "required_token_positions_per_second": 19005.90084627462,
+      "required_pages_per_second_upper_estimate": 0.0043966806683550744
     },
     {
       "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": 28343.66145317643,
-      "required_pages_per_second_upper_estimate": 0.0017259802747995364
+      "required_token_positions_per_second": 28508.851269411927,
+      "required_pages_per_second_upper_estimate": 0.0017360394679242933
     },
     {
       "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": 28343.66145317643,
-      "required_pages_per_second_upper_estimate": 0.0021819735189421052
+      "required_token_positions_per_second": 28508.851269411927,
+      "required_pages_per_second_upper_estimate": 0.002194690288270592
     },
     {
       "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": 28343.66145317643,
-      "required_pages_per_second_upper_estimate": 0.351933415115741
+      "required_token_positions_per_second": 28508.851269411927,
+      "required_pages_per_second_upper_estimate": 0.353984519778635
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.514666716562715,
-    "producer_p95_seconds": 0.37206209899886744,
-    "gpu_update_seconds": 0.9754561895849038,
-    "producer_batches_per_second": 1.9430049929761573,
-    "gpu_batches_per_second": 1.025161366217319,
-    "producer_headroom": 1.8953162467929652,
+    "producer_batch_seconds": 0.11019878539849515,
+    "producer_p95_seconds": 0.3683917200032738,
+    "gpu_update_seconds": 0.9698040702770945,
+    "producer_batches_per_second": 9.074510180705275,
+    "gpu_batches_per_second": 1.03113611362167,
+    "producer_headroom": 8.800496909018907,
     "sustainable": true,
-    "buffer_seconds": 60.478283754264034,
+    "buffer_seconds": 60.127852357179854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285246.770832637,
-    "upload_bytes_per_second": 5314139.929690056,
+    "generation_bytes_per_second": 285992.9229245364,
+    "upload_bytes_per_second": 5299938.032573958,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.2073168157037,
+    "estimated_full_shards_per_hour": 61.367423685093584,
     "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": 1.1647725229995558,
-      "bytes": 2632325,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.06261416399866,
       "bytes": 5980251,
@@ -562,6 +556,12 @@
       "bytes": 12102958,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.048562693002168,
+      "bytes": 883869,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 388400, "loss": 0.5341513473540545, "lr": 0.0003, "grad_norm": 0.12798725855584206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56851283147, "tokens_per_second": 148737.1506964806, "data_wait_s": 1.095100014936179e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388500, "loss": 0.578675290569663, "lr": 0.0003, "grad_norm": 0.1965254138468388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56865886657, "tokens_per_second": 140707.79495045563, "data_wait_s": 1.6524005332030356e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388600, "loss": 0.6301270946860313, "lr": 0.0003, "grad_norm": 0.1588578384734865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56880560507, "tokens_per_second": 150234.76658120236, "data_wait_s": 1.1785996321123093e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 388700, "loss": 0.6117727477103472, "lr": 0.0003, "grad_norm": 0.16343420983775175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56895205001, "tokens_per_second": 149016.2557481882, "data_wait_s": 1.4340999769046903e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:734d76a53c31ed0527192676137b4e1d338f73dd9be88617bada7ab7addc8eba
-size 716511
+oid sha256:59bc4f98d699ac95b2d39458a5d5010bb9fd4706e46c696ae518ce14e8dd8b64
+size 718629