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": 76535.96748669099,
-      "required_pages_per_second_upper_estimate": 0.0052949358186406914
+      "required_token_positions_per_second": 77062.59939900077,
+      "required_pages_per_second_upper_estimate": 0.005331369436288671
     },
     {
       "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": 9566.995935836374,
-      "required_pages_per_second_upper_estimate": 0.004005501416741627
+      "required_token_positions_per_second": 9632.824924875096,
+      "required_pages_per_second_upper_estimate": 0.004033062639786531
     },
     {
       "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": 19133.991871672748,
-      "required_pages_per_second_upper_estimate": 0.003985105909571079
+      "required_token_positions_per_second": 19265.649849750193,
+      "required_pages_per_second_upper_estimate": 0.004012526794350243
     },
     {
       "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": 28700.987807509122,
-      "required_pages_per_second_upper_estimate": 0.0017205803951892743
+      "required_token_positions_per_second": 28898.47477462529,
+      "required_pages_per_second_upper_estimate": 0.0017324194373227499
     },
     {
       "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": 28700.987807509122,
-      "required_pages_per_second_upper_estimate": 0.002231375559730167
+      "required_token_positions_per_second": 28898.47477462529,
+      "required_pages_per_second_upper_estimate": 0.0022467293027700827
     },
     {
       "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": 28700.987807509122,
-      "required_pages_per_second_upper_estimate": 0.36814545487499034
+      "required_token_positions_per_second": 28898.47477462529,
+      "required_pages_per_second_upper_estimate": 0.370678605644172
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4212558262185553,
-    "producer_p95_seconds": 0.2947685629987973,
-    "gpu_update_seconds": 0.9633117921037677,
-    "producer_batches_per_second": 2.3738544080840356,
-    "gpu_batches_per_second": 1.0380854965100232,
-    "producer_headroom": 2.286761944044861,
+    "producer_batch_seconds": 0.10482879684366253,
+    "producer_p95_seconds": 0.30431826100175385,
+    "gpu_update_seconds": 0.9567286929715998,
+    "producer_batches_per_second": 9.539363515650761,
+    "gpu_batches_per_second": 1.045228398966482,
+    "producer_headroom": 9.126582788109518,
     "sustainable": true,
-    "buffer_seconds": 60.68864290253737,
+    "buffer_seconds": 60.273907657210785,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359822.7117653961,
-    "upload_bytes_per_second": 4914070.019538895,
+    "generation_bytes_per_second": 359767.0475083522,
+    "upload_bytes_per_second": 4890712.926132742,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.20957770081914,
+    "estimated_full_shards_per_hour": 77.1976334470551,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.2576819179994345,
-      "bytes": 11872557,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.5458234039979288,
       "bytes": 11492771,
@@ -562,6 +556,12 @@
       "bytes": 2567501,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2765829149975616,
+      "bytes": 9916034,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -282,3 +282,4 @@
 {"step": 788200, "loss": 0.5107440948486328, "lr": 3e-05, "grad_norm": 0.2539574485578654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115370260563, "tokens_per_second": 153330.8123883191, "data_wait_s": 2.0318002498243004e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788300, "loss": 0.5174246430397034, "lr": 3e-05, "grad_norm": 0.2595425070103543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115384893956, "tokens_per_second": 152300.01003865947, "data_wait_s": 1.3340999430511147e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788400, "loss": 0.5367664098739624, "lr": 3e-05, "grad_norm": 0.27117663684232063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115399401173, "tokens_per_second": 150334.02912280656, "data_wait_s": 9.59799945121631e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 788500, "loss": 0.4945177435874939, "lr": 3e-05, "grad_norm": 0.25446166740222204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115413964021, "tokens_per_second": 151256.45725476622, "data_wait_s": 1.3527998817153275e-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 sha256:55f8caa8c69759d3f5bb1a3e80f4d486e7c047a4632eb40ee4d9205db493595d
-size 605592
+oid sha256:6ec8451129ffc5f629eccb9cd11688d736c8d02c76d767787c32b7909d51e36f
+size 607710