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": 77151.0195646952,
-      "required_pages_per_second_upper_estimate": 0.005337486548514462
+      "required_token_positions_per_second": 77488.2849876545,
+      "required_pages_per_second_upper_estimate": 0.005360819353038383
     },
     {
       "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": 9643.8774455869,
-      "required_pages_per_second_upper_estimate": 0.004037690099405685
+      "required_token_positions_per_second": 9686.035623456812,
+      "required_pages_per_second_upper_estimate": 0.0040553408481169536
     },
     {
       "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": 19287.7548911738,
-      "required_pages_per_second_upper_estimate": 0.004017130691529637
+      "required_token_positions_per_second": 19372.071246913623,
+      "required_pages_per_second_upper_estimate": 0.004034691564858412
     },
     {
       "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": 28931.6323367607,
-      "required_pages_per_second_upper_estimate": 0.0017344071825441017
+      "required_token_positions_per_second": 29058.106870370433,
+      "required_pages_per_second_upper_estimate": 0.0017419891377185742
     },
     {
       "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": 28931.6323367607,
-      "required_pages_per_second_upper_estimate": 0.0022493071573814086
+      "required_token_positions_per_second": 29058.106870370433,
+      "required_pages_per_second_upper_estimate": 0.0022591399960668838
     },
     {
       "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": 28931.6323367607,
-      "required_pages_per_second_upper_estimate": 0.37110391524942854
+      "required_token_positions_per_second": 29058.106870370433,
+      "required_pages_per_second_upper_estimate": 0.3727261947688002
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05437422121874924,
-    "producer_p95_seconds": 0.1335509739999452,
-    "gpu_update_seconds": 0.9556322186795624,
-    "producer_batches_per_second": 18.39106800218743,
-    "gpu_batches_per_second": 1.0464276742173286,
-    "producer_headroom": 17.57509711881708,
+    "producer_batch_seconds": 0.048050333992190275,
+    "producer_p95_seconds": 0.12089336699989417,
+    "gpu_update_seconds": 0.9514728582745949,
+    "producer_batches_per_second": 20.81150986718494,
+    "gpu_batches_per_second": 1.051002129281338,
+    "producer_headroom": 19.801586778340393,
     "sustainable": true,
-    "buffer_seconds": 60.20482977681243,
+    "buffer_seconds": 60.894262929574076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 30,
+    "upload_pages_per_commit": 38,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -1,2 +1,3 @@
 {"step": 600100, "loss": 0.6710267066955566, "lr": 0.0003, "grad_norm": 0.16837818088735748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87836984173, "tokens_per_second": 153948.01560736474, "data_wait_s": 1.1639000035756908e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600200, "loss": 0.5463886857032776, "lr": 0.0003, "grad_norm": 0.16265245312703575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87851654044, "tokens_per_second": 153987.49765921294, "data_wait_s": 1.036000003296067e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 600300, "loss": 0.5745486617088318, "lr": 0.0003, "grad_norm": 0.16983633367320333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87866176121, "tokens_per_second": 152576.3584327869, "data_wait_s": 8.928999932322768e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2481cd1fc3ff6008cf08bbd005ca99c9f4dc3302dd9ce855d48c9ddbac1397d2
-size 4401
+oid sha256:79d8b9b54cea0174792d87ed228044141378ea6158e4c2c602b0b697285f4c67
+size 6519