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": 76596.1878271695,
-      "required_pages_per_second_upper_estimate": 0.005299102001525436
+      "required_token_positions_per_second": 76875.12005014786,
+      "required_pages_per_second_upper_estimate": 0.005318399179923004
     },
     {
       "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": 9574.523478396188,
-      "required_pages_per_second_upper_estimate": 0.004008653041618458
+      "required_token_positions_per_second": 9609.390006268482,
+      "required_pages_per_second_upper_estimate": 0.004023250928742691
     },
     {
       "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": 19149.046956792376,
-      "required_pages_per_second_upper_estimate": 0.003988241486772203
+      "required_token_positions_per_second": 19218.780012536965,
+      "required_pages_per_second_upper_estimate": 0.004002765043299455
     },
     {
       "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": 28723.57043518856,
-      "required_pages_per_second_upper_estimate": 0.0017219341892369807
+      "required_token_positions_per_second": 28828.17001880545,
+      "required_pages_per_second_upper_estimate": 0.0017282047745604968
     },
     {
       "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": 28723.57043518856,
-      "required_pages_per_second_upper_estimate": 0.0022331312597017616
+      "required_token_positions_per_second": 28828.17001880545,
+      "required_pages_per_second_upper_estimate": 0.00224126341724303
     },
     {
       "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": 28723.57043518856,
-      "required_pages_per_second_upper_estimate": 0.36843512057552574
+      "required_token_positions_per_second": 28828.17001880545,
+      "required_pages_per_second_upper_estimate": 0.3697768117238805
     }
   ],
   "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": 0.1190002645784034,
-    "producer_p95_seconds": 0.33433549199980916,
-    "gpu_update_seconds": 0.9625544311207598,
-    "producer_batches_per_second": 8.403342660983323,
-    "gpu_batches_per_second": 1.038902287152364,
-    "producer_headroom": 8.088674714555614,
+    "producer_batch_seconds": 0.15875891118747631,
+    "producer_p95_seconds": 0.33013940200180514,
+    "gpu_update_seconds": 0.959061916936261,
+    "producer_batches_per_second": 6.298859021646433,
+    "gpu_batches_per_second": 1.042685547555174,
+    "producer_headroom": 6.040995807811489,
     "sustainable": true,
-    "buffer_seconds": 60.64092916060787,
+    "buffer_seconds": 60.42090076698444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442165.04628718283,
-    "upload_bytes_per_second": 2957775.588587294,
+    "generation_bytes_per_second": 455554.50848697673,
+    "upload_bytes_per_second": 2926869.8237313447,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.87832585775007,
+    "estimated_full_shards_per_hour": 97.75139275509812,
     "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": 1.3758144730018103,
-      "bytes": 6889142,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.881114518000686,
       "bytes": 7983801,
@@ -562,6 +556,12 @@
       "bytes": 6058293,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.5040729690008448,
+      "bytes": 2768823,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 884500, "loss": 0.5481231212615967, "lr": 3e-05, "grad_norm": 0.31447723963001145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129466019236, "tokens_per_second": 153058.77693741233, "data_wait_s": 1.3557997590396553e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884600, "loss": 0.5369524955749512, "lr": 3e-05, "grad_norm": 0.30164804543223034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129480555436, "tokens_per_second": 151341.2892795568, "data_wait_s": 1.3429002137854695e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884700, "loss": 0.5245838165283203, "lr": 3e-05, "grad_norm": 0.31288015655923224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129495334134, "tokens_per_second": 152868.24726632424, "data_wait_s": 1.142800465459004e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 884800, "loss": 0.5430890321731567, "lr": 3e-05, "grad_norm": 0.31524290159184143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129510108848, "tokens_per_second": 153653.34742957065, "data_wait_s": 1.9150000298395753e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d13b5987231338c546cdff71ebc3fc677c32ebde27a01dae6df24cc5ec3b1f5
-size 900225
+oid sha256:c71406a50c0ee2a6358487b9748d67aae61c3d2d3fae76b216b1d391b59b0aec
+size 902343