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": 75888.42319126465,
-      "required_pages_per_second_upper_estimate": 0.006577816184768676
+      "required_token_positions_per_second": 76297.00497289494,
+      "required_pages_per_second_upper_estimate": 0.006613231018059332
     },
     {
       "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": 9486.052898908081,
-      "required_pages_per_second_upper_estimate": 0.003784128330504261
+      "required_token_positions_per_second": 9537.125621611867,
+      "required_pages_per_second_upper_estimate": 0.0038045020031960536
     },
     {
       "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": 18972.105797816163,
-      "required_pages_per_second_upper_estimate": 0.004388862778666749
+      "required_token_positions_per_second": 19074.251243223734,
+      "required_pages_per_second_upper_estimate": 0.00441249232976335
     },
     {
       "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": 28458.158696724244,
-      "required_pages_per_second_upper_estimate": 0.0017329525562109875
+      "required_token_positions_per_second": 28611.376864835605,
+      "required_pages_per_second_upper_estimate": 0.0017422827387753679
     },
     {
       "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": 28458.158696724244,
-      "required_pages_per_second_upper_estimate": 0.0021907878336990716
+      "required_token_positions_per_second": 28611.376864835605,
+      "required_pages_per_second_upper_estimate": 0.0022025829924153194
     },
     {
       "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": 28458.158696724244,
-      "required_pages_per_second_upper_estimate": 0.353355087682981
+      "required_token_positions_per_second": 28611.376864835605,
+      "required_pages_per_second_upper_estimate": 0.3552575445427022
     }
   ],
   "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": 0.0460616731875092,
-    "producer_p95_seconds": 0.11525120300007075,
-    "gpu_update_seconds": 0.9715315841281924,
-    "producer_batches_per_second": 21.710023340428187,
-    "gpu_batches_per_second": 1.0293026148988804,
-    "producer_headroom": 21.091973367386228,
+    "producer_batch_seconds": 0.047923634468755694,
+    "producer_p95_seconds": 0.11953443099992,
+    "gpu_update_seconds": 0.9663288883514156,
+    "producer_batches_per_second": 20.86653091079643,
+    "gpu_batches_per_second": 1.0348443599839265,
+    "producer_headroom": 20.163931618780367,
     "sustainable": true,
-    "buffer_seconds": 60.23495821594793,
+    "buffer_seconds": 60.87871996613919,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 47,
+    "upload_pages_per_commit": 51,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -2,3 +2,4 @@
 {"step": 275200, "loss": 0.5517940912395716, "lr": 0.0003, "grad_norm": 0.1433728997387089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40280693026, "tokens_per_second": 149900.41362321944, "data_wait_s": 1.509800017629459e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275300, "loss": 0.5715921688824892, "lr": 0.0003, "grad_norm": 0.13326310819662313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40295403551, "tokens_per_second": 151598.20565482703, "data_wait_s": 1.600599989615148e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275400, "loss": 0.584355790168047, "lr": 0.0003, "grad_norm": 0.15548764254166622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40310118451, "tokens_per_second": 151720.85146898538, "data_wait_s": 1.5663000112908776e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275500, "loss": 0.6548275984823704, "lr": 0.0003, "grad_norm": 0.15053094391066363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40324827474, "tokens_per_second": 151467.69523299168, "data_wait_s": 1.3049000017417711e-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 sha256:03b5fb02236c87cc9c6d118ca8d6188f743206b3ac207b6a413f02f0ce7c2e59
-size 8637
+oid sha256:c3e0a168de91e3d5a070237149493535ec4864fc4ae535e936667b4761977db6
+size 10755