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": 76016.34410052048,
-      "required_pages_per_second_upper_estimate": 0.006588904044970391
+      "required_token_positions_per_second": 76581.19985415826,
+      "required_pages_per_second_upper_estimate": 0.0066378643100818945
     },
     {
       "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": 9502.04301256506,
-      "required_pages_per_second_upper_estimate": 0.003790507025915534
+      "required_token_positions_per_second": 9572.649981769782,
+      "required_pages_per_second_upper_estimate": 0.003818673201599562
     },
     {
       "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": 19004.08602513012,
-      "required_pages_per_second_upper_estimate": 0.004396260841423055
+      "required_token_positions_per_second": 19145.299963539564,
+      "required_pages_per_second_upper_estimate": 0.004428928200793647
     },
     {
       "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": 28506.12903769518,
-      "required_pages_per_second_upper_estimate": 0.0017358736982951892
+      "required_token_positions_per_second": 28717.949945309345,
+      "required_pages_per_second_upper_estimate": 0.0017487724802304784
     },
     {
       "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": 28506.12903769518,
-      "required_pages_per_second_upper_estimate": 0.002194480723337417
+      "required_token_positions_per_second": 28717.949945309345,
+      "required_pages_per_second_upper_estimate": 0.002210787283163356
     },
     {
       "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": 28506.12903769518,
-      "required_pages_per_second_upper_estimate": 0.35395071877143647
+      "required_token_positions_per_second": 28717.949945309345,
+      "required_pages_per_second_upper_estimate": 0.35658082552503007
     }
   ],
   "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": 0.05591853085939036,
-    "producer_p95_seconds": 0.1283456710000337,
-    "gpu_update_seconds": 0.9698966830410249,
-    "producer_batches_per_second": 17.883159386188893,
-    "gpu_batches_per_second": 1.0310376532731185,
-    "producer_headroom": 17.34481697095858,
+    "producer_batch_seconds": 0.048221036781244564,
+    "producer_p95_seconds": 0.11981866199994329,
+    "gpu_update_seconds": 0.9627428159967211,
+    "producer_batches_per_second": 20.737836984644577,
+    "gpu_batches_per_second": 1.0386989997580058,
+    "producer_headroom": 19.96520357627767,
     "sustainable": true,
-    "buffer_seconds": 60.13359434854355,
+    "buffer_seconds": 60.65279740779343,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 51,
+    "upload_pages_per_commit": 54,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -4,3 +4,4 @@
 {"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}
 {"step": 275600, "loss": 0.6014189012348652, "lr": 0.0003, "grad_norm": 0.14295170022911294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40339605846, "tokens_per_second": 152758.6646086566, "data_wait_s": 1.4557999975295388e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275700, "loss": 0.5936791710555553, "lr": 0.0003, "grad_norm": 0.15593649290911388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40354109564, "tokens_per_second": 149561.37570848665, "data_wait_s": 1.07790001493413e-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:a644fe477251c6a931054bc514f3cfa99a428d7de3dfe669f3f1fe4d7db224fe
-size 12873
+oid sha256:439e6f9935a1e755e888a35d37c9cc918db964e4ce6080d9581d6747150edbc4
+size 14991