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": 75123.39200845319,
-      "required_pages_per_second_upper_estimate": 0.0065115052207963845
+      "required_token_positions_per_second": 75892.59908103506,
+      "required_pages_per_second_upper_estimate": 0.0065781781403102795
     },
     {
       "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": 9390.424001056648,
-      "required_pages_per_second_upper_estimate": 0.0037459805333718878
+      "required_token_positions_per_second": 9486.574885129383,
+      "required_pages_per_second_upper_estimate": 0.0037843365586123275
     },
     {
       "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": 18780.848002113296,
-      "required_pages_per_second_upper_estimate": 0.004344618653653131
+      "required_token_positions_per_second": 18973.149770258766,
+      "required_pages_per_second_upper_estimate": 0.004389104283318058
     },
     {
       "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": 28171.272003169946,
-      "required_pages_per_second_upper_estimate": 0.0017154826617516855
+      "required_token_positions_per_second": 28459.72465538815,
+      "required_pages_per_second_upper_estimate": 0.001733047914877669
     },
     {
       "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": 28171.272003169946,
-      "required_pages_per_second_upper_estimate": 0.002168702501876066
+      "required_token_positions_per_second": 28459.72465538815,
+      "required_pages_per_second_upper_estimate": 0.00219090838553187
     },
     {
       "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": 28171.272003169946,
-      "required_pages_per_second_upper_estimate": 0.3497929150970355
+      "required_token_positions_per_second": 28459.72465538815,
+      "required_pages_per_second_upper_estimate": 0.35337453164866023
     }
   ],
   "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": 0.14308825507799838,
-    "producer_p95_seconds": 0.3772982719965512,
-    "gpu_update_seconds": 0.9814253327605844,
-    "producer_batches_per_second": 6.988693792197642,
-    "gpu_batches_per_second": 1.0189262153924314,
-    "producer_headroom": 6.858881130569401,
+    "producer_batch_seconds": 0.22836194517998365,
+    "producer_p95_seconds": 0.38241851199563826,
+    "gpu_update_seconds": 0.97147812688924,
+    "producer_batches_per_second": 4.379013321207477,
+    "gpu_batches_per_second": 1.029359254028796,
+    "producer_headroom": 4.25411565890967,
     "sustainable": true,
-    "buffer_seconds": 60.848370631156236,
+    "buffer_seconds": 60.231643867132874,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435124.21908400964,
-    "upload_bytes_per_second": 5150203.197842307,
+    "generation_bytes_per_second": 407597.1286645371,
+    "upload_bytes_per_second": 5224545.040174899,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.36752824201791,
+    "estimated_full_shards_per_hour": 87.46085543586811,
     "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": 1.054510723999556,
-      "bytes": 1093266,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.209129931001371,
       "bytes": 7450282,
@@ -562,6 +556,12 @@
       "bytes": 5714080,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1845541919974494,
+      "bytes": 7669056,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -373,3 +373,4 @@
 {"step": 392300, "loss": 0.6390158757567406, "lr": 0.0003, "grad_norm": 0.19642958353865456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57422109078, "tokens_per_second": 149970.703533264, "data_wait_s": 1.0579999070614576e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 392400, "loss": 0.5793743394315243, "lr": 0.0003, "grad_norm": 0.1551709794784521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57436752854, "tokens_per_second": 149779.14548028368, "data_wait_s": 1.0162002581637353e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 392500, "loss": 0.5946585964411497, "lr": 0.0003, "grad_norm": 0.18924079779568195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57451452603, "tokens_per_second": 150697.16982005257, "data_wait_s": 1.5876001270953566e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 392600, "loss": 0.5372662991285324, "lr": 0.0003, "grad_norm": 0.15039931475308008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57466104935, "tokens_per_second": 146378.7179142551, "data_wait_s": 9.699004294816405e-06, "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 sha256:37f698b0dbdd49925ec0f20b11bcd746ee1fde9587d1639753e29a57cdeae47f
-size 799896
+oid sha256:407b07f7083b12f02494346d4078174815c4fd96c4e07b242ce7744c37cca605
+size 802014