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": 77273.43590364498,
-      "required_pages_per_second_upper_estimate": 0.005345955594888047
+      "required_token_positions_per_second": 76744.23066109727,
+      "required_pages_per_second_upper_estimate": 0.005309343948283259
     },
     {
       "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": 9659.179487955622,
-      "required_pages_per_second_upper_estimate": 0.004044096745002488
+      "required_token_positions_per_second": 9593.02883263716,
+      "required_pages_per_second_upper_estimate": 0.0040164008470034126
     },
     {
       "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": 19318.358975911244,
-      "required_pages_per_second_upper_estimate": 0.004023504715296457
+      "required_token_positions_per_second": 19186.05766527432,
+      "required_pages_per_second_upper_estimate": 0.0039959498413109735
     },
     {
       "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": 28977.538463866866,
-      "required_pages_per_second_upper_estimate": 0.0017371591847695188
+      "required_token_positions_per_second": 28779.086497911478,
+      "required_pages_per_second_upper_estimate": 0.0017252622924291
     },
     {
       "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": 28977.538463866866,
-      "required_pages_per_second_upper_estimate": 0.002252876156844199
+      "required_token_positions_per_second": 28779.086497911478,
+      "required_pages_per_second_upper_estimate": 0.002237447389389115
     },
     {
       "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": 28977.538463866866,
-      "required_pages_per_second_upper_estimate": 0.37169274975778743
+      "required_token_positions_per_second": 28779.086497911478,
+      "required_pages_per_second_upper_estimate": 0.3691472210196313
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11104221257039626,
-    "producer_p95_seconds": 0.1834554830002162,
-    "gpu_update_seconds": 0.9541183090646325,
-    "producer_batches_per_second": 9.005584244514585,
-    "gpu_batches_per_second": 1.0480880520785179,
-    "producer_headroom": 8.592392811515351,
+    "producer_batch_seconds": 0.5173437683438067,
+    "producer_p95_seconds": 0.21470924200002628,
+    "gpu_update_seconds": 0.9606976233247166,
+    "producer_batches_per_second": 1.9329507016221341,
+    "gpu_batches_per_second": 1.040910246596914,
+    "producer_headroom": 1.8569811450522278,
     "sustainable": true,
-    "buffer_seconds": 60.109453471071845,
+    "buffer_seconds": 60.523950269457146,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327336.3343132018,
-    "upload_bytes_per_second": 4617155.832011391,
+    "generation_bytes_per_second": 295374.36005303677,
+    "upload_bytes_per_second": 4562604.5339809945,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.23875734374084,
+    "estimated_full_shards_per_hour": 63.380461704190516,
     "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": 1.1896957620001558,
-      "bytes": 10758587,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 0.9081728179999118,
       "bytes": 2358450,
@@ -562,6 +556,12 @@
       "bytes": 257606,
       "pages": 8,
       "shards": 1
+    },
+    {
+      "seconds": 0.9524619530002383,
+      "bytes": 4638620,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 535000, "loss": 0.6051273345947266, "lr": 0.0003, "grad_norm": 0.16704291770607504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78306858501, "tokens_per_second": 153400.40848717812, "data_wait_s": 1.5859999621170573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 535100, "loss": 0.6086952090263367, "lr": 0.0003, "grad_norm": 0.16169766829009985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78321537261, "tokens_per_second": 146518.9366638648, "data_wait_s": 8.11400059319567e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 535200, "loss": 0.585863471031189, "lr": 0.0003, "grad_norm": 0.16026792908453769, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78336244560, "tokens_per_second": 147403.01766358546, "data_wait_s": 1.7534999642521143e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 535300, "loss": 0.6077885031700134, "lr": 0.0003, "grad_norm": 0.1599695314387667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78350724100, "tokens_per_second": 151344.8475798024, "data_wait_s": 1.4061999536352232e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed91ab40109b11f6f7cf4f7dd169538990a39ec03ede34e58d157d1673fe952b
-size 217767
+oid sha256:bbb3d5216c46e2d6bc3f2172e0bbdb4bc98ffd8f402fe037673d303f8736ac83
+size 219885