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": 77493.1983603359,
-      "required_pages_per_second_upper_estimate": 0.005361159271561078
+      "required_token_positions_per_second": 77090.20554692595,
+      "required_pages_per_second_upper_estimate": 0.005333279293657223
     },
     {
       "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": 9686.649795041987,
-      "required_pages_per_second_upper_estimate": 0.004055597988934305
+      "required_token_positions_per_second": 9636.275693365744,
+      "required_pages_per_second_upper_estimate": 0.004034507404493325
     },
     {
       "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": 19373.299590083974,
-      "required_pages_per_second_upper_estimate": 0.004034947396347208
+      "required_token_positions_per_second": 19272.551386731488,
+      "required_pages_per_second_upper_estimate": 0.0040139642024976776
     },
     {
       "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": 29059.949385125965,
-      "required_pages_per_second_upper_estimate": 0.001742099593664811
+      "required_token_positions_per_second": 28908.82708009723,
+      "required_pages_per_second_upper_estimate": 0.0017330400422287374
     },
     {
       "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": 29059.949385125965,
-      "required_pages_per_second_upper_estimate": 0.0022592832434847605
+      "required_token_positions_per_second": 28908.82708009723,
+      "required_pages_per_second_upper_estimate": 0.002247534148985547
     },
     {
       "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": 29059.949385125965,
-      "required_pages_per_second_upper_estimate": 0.372749828569746
+      "required_token_positions_per_second": 28908.82708009723,
+      "required_pages_per_second_upper_estimate": 0.3708113939033264
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0506279414453239,
-    "producer_p95_seconds": 0.1350936300000285,
-    "gpu_update_seconds": 0.9514125311639856,
-    "producer_batches_per_second": 19.751938780287148,
-    "gpu_batches_per_second": 1.0510687711634101,
-    "producer_headroom": 18.79224207034908,
+    "producer_batch_seconds": 0.04636835622655866,
+    "producer_p95_seconds": 0.12498815499998273,
+    "gpu_update_seconds": 0.956386086623166,
+    "producer_batches_per_second": 21.566431967394706,
+    "gpu_batches_per_second": 1.0456028313113872,
+    "producer_headroom": 20.62583547172137,
     "sustainable": true,
-    "buffer_seconds": 60.89040199449508,
+    "buffer_seconds": 60.25232345725946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 66,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 761300, "loss": 0.567022442817688, "lr": 3e-05, "grad_norm": 0.22644278490856187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111432132086, "tokens_per_second": 154186.37125312866, "data_wait_s": 9.705999900688767e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 761400, "loss": 0.5207698345184326, "lr": 3e-05, "grad_norm": 0.22618513057158895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111446784082, "tokens_per_second": 153802.0080439591, "data_wait_s": 9.147000127995852e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 761500, "loss": 0.5549948215484619, "lr": 3e-05, "grad_norm": 0.25229013580404586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111461313256, "tokens_per_second": 152066.6605562098, "data_wait_s": 8.672999911141233e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 761600, "loss": 0.5470552444458008, "lr": 3e-05, "grad_norm": 0.25005745380968136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111475755654, "tokens_per_second": 151434.22608220068, "data_wait_s": 1.94229999124218e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd75e541e6fde82f11d4b3ced120160b972c8d6664562d76c89ce04441205868
-size 31935
+oid sha256:57da94bf8dad0adf5ea3cd208059ad986d6a0e027ffc53ec7915e2864b0889fe
+size 34053