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": 76756.00109446014,
-      "required_pages_per_second_upper_estimate": 0.005310158254174471
+      "required_token_positions_per_second": 76812.83564387822,
+      "required_pages_per_second_upper_estimate": 0.005314090200177537
     },
     {
       "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": 9594.500136807517,
-      "required_pages_per_second_upper_estimate": 0.004017016851335217
+      "required_token_positions_per_second": 9601.604455484778,
+      "required_pages_per_second_upper_estimate": 0.004019991281210342
     },
     {
       "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": 19189.000273615035,
-      "required_pages_per_second_upper_estimate": 0.0039965627090265445
+      "required_token_positions_per_second": 19203.208910969555,
+      "required_pages_per_second_upper_estimate": 0.003999521993480526
     },
     {
       "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": 28783.500410422555,
-      "required_pages_per_second_upper_estimate": 0.0017255268997444849
+      "required_token_positions_per_second": 28804.813366454335,
+      "required_pages_per_second_upper_estimate": 0.0017268045789155906
     },
     {
       "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": 28783.500410422555,
-      "required_pages_per_second_upper_estimate": 0.0022377905516721024
+      "required_token_positions_per_second": 28804.813366454335,
+      "required_pages_per_second_upper_estimate": 0.0022394475402577867
     },
     {
       "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": 28783.500410422555,
-      "required_pages_per_second_upper_estimate": 0.3692038379500334
+      "required_token_positions_per_second": 28804.813366454335,
+      "required_pages_per_second_upper_estimate": 0.3694772176659398
     }
   ],
   "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": 0.5067812572185346,
-    "producer_p95_seconds": 0.31933896799819195,
-    "gpu_update_seconds": 0.9605503015883576,
-    "producer_batches_per_second": 1.9732379320586815,
-    "gpu_batches_per_second": 1.041069893316788,
-    "producer_headroom": 1.8953942907445536,
+    "producer_batch_seconds": 0.5553132609608156,
+    "producer_p95_seconds": 0.3167180290001852,
+    "gpu_update_seconds": 0.9598395812624309,
+    "producer_batches_per_second": 1.800785377013647,
+    "gpu_batches_per_second": 1.0418407612288172,
+    "producer_headroom": 1.7284650822162877,
     "sustainable": true,
-    "buffer_seconds": 60.51466900006653,
+    "buffer_seconds": 60.469893619533146,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285083.5321892825,
-    "upload_bytes_per_second": 3137649.211734167,
+    "generation_bytes_per_second": 247522.86638889826,
+    "upload_bytes_per_second": 3195819.317741498,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.172289602840955,
+    "estimated_full_shards_per_hour": 53.11264508962832,
     "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": 1.7772512390001793,
-      "bytes": 2617112,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.5108418660020106,
       "bytes": 11046966,
@@ -562,6 +556,12 @@
       "bytes": 2366813,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.8487111699978414,
+      "bytes": 10763047,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 831500, "loss": 0.5391645431518555, "lr": 3e-05, "grad_norm": 0.29102243099453995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121708218681, "tokens_per_second": 151881.14229352007, "data_wait_s": 1.3131000741850585e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 831600, "loss": 0.4987949728965759, "lr": 3e-05, "grad_norm": 0.2921272281507087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121722834670, "tokens_per_second": 150944.01107691918, "data_wait_s": 1.5281999367289245e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 831700, "loss": 0.5129266977310181, "lr": 3e-05, "grad_norm": 0.29703569636226895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121737547254, "tokens_per_second": 129535.01823401765, "data_wait_s": 1.7319001926807687e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 831800, "loss": 0.5605382919311523, "lr": 3e-05, "grad_norm": 0.2931005483054918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121752186886, "tokens_per_second": 151140.40237385005, "data_wait_s": 9.47800072026439e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2894ec18fdc14bdcd5a538535a3411fff1b17870fbc443b3fb0549f81407ffd3
-size 676269
+oid sha256:2fce9c5bdcb7d0848ed6cfd46134b0e8332839315cc501d9779b17178a131fb2
+size 678387