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": 76286.93470903172,
-      "required_pages_per_second_upper_estimate": 0.005277707153246569
+      "required_token_positions_per_second": 76396.5213561999,
+      "required_pages_per_second_upper_estimate": 0.005285288611773717
     },
     {
       "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": 9535.866838628965,
-      "required_pages_per_second_upper_estimate": 0.003992468313790354
+      "required_token_positions_per_second": 9549.565169524987,
+      "required_pages_per_second_upper_estimate": 0.003998203518882842
     },
     {
       "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": 19071.73367725793,
-      "required_pages_per_second_upper_estimate": 0.003972139169533469
+      "required_token_positions_per_second": 19099.130339049974,
+      "required_pages_per_second_upper_estimate": 0.003977845171686195
     },
     {
       "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": 28607.600515886894,
-      "required_pages_per_second_upper_estimate": 0.0017149819696506594
+      "required_token_positions_per_second": 28648.695508574965,
+      "required_pages_per_second_upper_estimate": 0.0017174455517138887
     },
     {
       "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": 28607.600515886894,
-      "required_pages_per_second_upper_estimate": 0.002224115108573823
+      "required_token_positions_per_second": 28648.695508574965,
+      "required_pages_per_second_upper_estimate": 0.002227310063497551
     },
     {
       "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": 28607.600515886894,
-      "required_pages_per_second_upper_estimate": 0.36694758296952185
+      "required_token_positions_per_second": 28648.695508574965,
+      "required_pages_per_second_upper_estimate": 0.36747470541135907
     }
   ],
   "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": 0.09194111444503505,
-    "producer_p95_seconds": 0.32057406900275964,
-    "gpu_update_seconds": 0.9664564486855866,
-    "producer_batches_per_second": 10.876526851301414,
-    "gpu_batches_per_second": 1.034707773288733,
-    "producer_headroom": 10.511689514742189,
+    "producer_batch_seconds": 0.46940194996852824,
+    "producer_p95_seconds": 0.3268520170022384,
+    "gpu_update_seconds": 0.9650701195705249,
+    "producer_batches_per_second": 2.130370357573177,
+    "gpu_batches_per_second": 1.0361941373182495,
+    "producer_headroom": 2.0559567757126476,
     "sustainable": true,
-    "buffer_seconds": 60.886756267191956,
+    "buffer_seconds": 60.79941753294307,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285857.96787886065,
-    "upload_bytes_per_second": 4858869.078796342,
+    "generation_bytes_per_second": 291362.35642221745,
+    "upload_bytes_per_second": 4838105.75540349,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.33846547388425,
+    "estimated_full_shards_per_hour": 62.51957911968129,
     "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": 1.1101431559982302,
-      "bytes": 7100100,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0310441840010753,
       "bytes": 5933626,
@@ -562,6 +556,12 @@
       "bytes": 5970136,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2014365330032888,
+      "bytes": 5687823,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -382,3 +382,4 @@
 {"step": 758200, "loss": 0.5575207471847534, "lr": 5e-05, "grad_norm": 0.24727222334973173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110979052216, "tokens_per_second": 151033.8948268855, "data_wait_s": 9.016999683808535e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 758300, "loss": 0.5599696636199951, "lr": 5e-05, "grad_norm": 0.2452968377925745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110993613847, "tokens_per_second": 150560.9622968602, "data_wait_s": 1.1952004570048302e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758400, "loss": 0.5410264134407043, "lr": 5e-05, "grad_norm": 0.2684164749589962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111008130084, "tokens_per_second": 149818.40943257354, "data_wait_s": 1.3881006452720612e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 758500, "loss": 0.5073688626289368, "lr": 5e-05, "grad_norm": 0.2388039053231994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111022668129, "tokens_per_second": 150572.55613309293, "data_wait_s": 1.0691997886169702e-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:c8783c14f0cd2ce93608e0aa5269bee4d08ed94fc0eef7d0e1c7083327692304
-size 818958
+oid sha256:82ce59d777a842e6f29e0ca085804907f0de5d6744021b2bc997ba0564652da0
+size 821076