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": 76789.85829493667,
-      "required_pages_per_second_upper_estimate": 0.005312500573862965
+      "required_token_positions_per_second": 76396.56749591012,
+      "required_pages_per_second_upper_estimate": 0.005285291803825929
     },
     {
       "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": 9598.732286867084,
-      "required_pages_per_second_upper_estimate": 0.004018788764187814
+      "required_token_positions_per_second": 9549.570936988765,
+      "required_pages_per_second_upper_estimate": 0.003998205933599487
     },
     {
       "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": 19197.464573734167,
-      "required_pages_per_second_upper_estimate": 0.003998325599522755
+      "required_token_positions_per_second": 19099.14187397753,
+      "required_pages_per_second_upper_estimate": 0.003977847574107408
     },
     {
       "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": 28796.19686060125,
-      "required_pages_per_second_upper_estimate": 0.0017262880325463407
+      "required_token_positions_per_second": 28648.71281096629,
+      "required_pages_per_second_upper_estimate": 0.0017174465889658295
     },
     {
       "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": 28796.19686060125,
-      "required_pages_per_second_upper_estimate": 0.0022387776448277133
+      "required_token_positions_per_second": 28648.71281096629,
+      "required_pages_per_second_upper_estimate": 0.0022273114086823714
     },
     {
       "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": 28796.19686060125,
-      "required_pages_per_second_upper_estimate": 0.3693666943805396
+      "required_token_positions_per_second": 28648.71281096629,
+      "required_pages_per_second_upper_estimate": 0.36747492734785714
     }
   ],
   "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": 0.08754691208568488,
-    "producer_p95_seconds": 0.25593282599948,
-    "gpu_update_seconds": 0.9601267880560921,
-    "producer_batches_per_second": 11.42244741906224,
-    "gpu_batches_per_second": 1.0415291109881817,
-    "producer_headroom": 10.966997752203826,
+    "producer_batch_seconds": 0.10753726667977048,
+    "producer_p95_seconds": 0.25298658500105375,
+    "gpu_update_seconds": 0.9650695367163848,
+    "producer_batches_per_second": 9.29910189160607,
+    "gpu_batches_per_second": 1.0361947631281212,
+    "producer_headroom": 8.974279954410727,
     "sustainable": true,
-    "buffer_seconds": 60.4879876475338,
+    "buffer_seconds": 60.799380813132245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344653.14986987703,
-    "upload_bytes_per_second": 5373950.195730494,
+    "generation_bytes_per_second": 341495.7370970572,
+    "upload_bytes_per_second": 5387116.212514579,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.95454284736856,
+    "estimated_full_shards_per_hour": 73.27703556712902,
     "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": 1.1375966550003795,
-      "bytes": 6241738,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.329601728997659,
       "bytes": 12006565,
@@ -562,6 +556,12 @@
       "bytes": 6807685,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1573900719995436,
+      "bytes": 7406360,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 745300, "loss": 0.5240442156791687, "lr": 5e-05, "grad_norm": 0.23911306087253575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109091459873, "tokens_per_second": 149986.14113557548, "data_wait_s": 1.0012001439463347e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745400, "loss": 0.5447136759757996, "lr": 5e-05, "grad_norm": 0.22656630798232283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109105958134, "tokens_per_second": 147603.6770643994, "data_wait_s": 1.3957996998215094e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745500, "loss": 0.5598964095115662, "lr": 5e-05, "grad_norm": 0.2301542085667018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109120644051, "tokens_per_second": 151447.52547444103, "data_wait_s": 1.556699862703681e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 745600, "loss": 0.5423048138618469, "lr": 5e-05, "grad_norm": 0.23486748478399427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109135360963, "tokens_per_second": 151800.3414922986, "data_wait_s": 1.3477998436428607e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3836f85809b6ffafeb5e981522ee069cddcbf1f4bb99182b35ca193e548c7eb
-size 544170
+oid sha256:54dec80053c59b2ea78cfb1e98eda8f000434fd803979c6db2315ecf318b0c0e
+size 546288