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": 75938.26074882413,
-      "required_pages_per_second_upper_estimate": 0.0065821359779457975
+      "required_token_positions_per_second": 76194.2268773001,
+      "required_pages_per_second_upper_estimate": 0.006604322473221872
     },
     {
       "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": 9492.282593603017,
-      "required_pages_per_second_upper_estimate": 0.003786613448860307
+      "required_token_positions_per_second": 9524.278359662512,
+      "required_pages_per_second_upper_estimate": 0.0037993770383207724
     },
     {
       "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": 18984.565187206033,
-      "required_pages_per_second_upper_estimate": 0.004391745039124331
+      "required_token_positions_per_second": 19048.556719325024,
+      "required_pages_per_second_upper_estimate": 0.004406548353867559
     },
     {
       "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": 28476.84778080905,
-      "required_pages_per_second_upper_estimate": 0.001734090623377706
+      "required_token_positions_per_second": 28572.835078987537,
+      "required_pages_per_second_upper_estimate": 0.0017399357462303429
     },
     {
       "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": 28476.84778080905,
-      "required_pages_per_second_upper_estimate": 0.002192226571126616
+      "required_token_positions_per_second": 28572.835078987537,
+      "required_pages_per_second_upper_estimate": 0.002199615939050243
     },
     {
       "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": 28476.84778080905,
-      "required_pages_per_second_upper_estimate": 0.3535871435589735
+      "required_token_positions_per_second": 28572.835078987537,
+      "required_pages_per_second_upper_estimate": 0.35477898455352863
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05404563364841408,
-    "producer_p95_seconds": 0.15243659799989473,
-    "gpu_update_seconds": 0.9708939771989924,
-    "producer_batches_per_second": 18.502882332832897,
-    "gpu_batches_per_second": 1.0299785800350496,
-    "producer_headroom": 17.9643370177691,
+    "producer_batch_seconds": 0.05446702034380024,
+    "producer_p95_seconds": 0.1516409770001701,
+    "gpu_update_seconds": 0.9676323656217208,
+    "producer_batches_per_second": 18.359733902973193,
+    "gpu_batches_per_second": 1.0334503428453248,
+    "producer_headroom": 17.76547274871926,
     "sustainable": true,
-    "buffer_seconds": 60.195426586337526,
+    "buffer_seconds": 60.960839034168416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 278300, "loss": 0.6166594307869673, "lr": 0.0003, "grad_norm": 0.17162817212766507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40735477528, "tokens_per_second": 151484.95224747804, "data_wait_s": 1.0196999937761575e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278400, "loss": 0.6390067264437675, "lr": 0.0003, "grad_norm": 0.164547309035477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40750257836, "tokens_per_second": 152601.1102714849, "data_wait_s": 1.526100004412001e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278500, "loss": 0.6061176229268312, "lr": 0.0003, "grad_norm": 0.16668151308511772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40765085288, "tokens_per_second": 152982.57103350223, "data_wait_s": 1.4629999895987567e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278600, "loss": 0.5726190768182278, "lr": 0.0003, "grad_norm": 0.13401794960571295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40779598508, "tokens_per_second": 149480.0919833245, "data_wait_s": 1.6554000467294827e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f348630c760aa27f75b2c4c9b55fd460fff5279314ba9693a4f4b27d2d52c4a5
-size 74295
+oid sha256:8207d1313e5f07a11d88d7f9374a29bbf60dfa48ad33b9e776e65e17ed154df6
+size 76413