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": 77043.90787572875,
-      "required_pages_per_second_upper_estimate": 0.005330076313338403
+      "required_token_positions_per_second": 77380.99936979407,
+      "required_pages_per_second_upper_estimate": 0.005353397085057869
     },
     {
       "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": 9630.488484466094,
-      "required_pages_per_second_upper_estimate": 0.004032084420977705
+      "required_token_positions_per_second": 9672.624921224258,
+      "required_pages_per_second_upper_estimate": 0.004049726067139491
     },
     {
       "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": 19260.976968932187,
-      "required_pages_per_second_upper_estimate": 0.0040115535565080065
+      "required_token_positions_per_second": 19345.249842448517,
+      "required_pages_per_second_upper_estimate": 0.004029105373636332
     },
     {
       "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": 28891.465453398283,
-      "required_pages_per_second_upper_estimate": 0.0017319992392178045
+      "required_token_positions_per_second": 29017.874763672775,
+      "required_pages_per_second_upper_estimate": 0.0017395772843529194
     },
     {
       "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": 28891.465453398283,
-      "required_pages_per_second_upper_estimate": 0.002246184358875429
+      "required_token_positions_per_second": 29017.874763672775,
+      "required_pages_per_second_upper_estimate": 0.002256012126733476
     },
     {
       "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": 28891.465453398283,
-      "required_pages_per_second_upper_estimate": 0.3705886975974947
+      "required_token_positions_per_second": 29017.874763672775,
+      "required_pages_per_second_upper_estimate": 0.3722101405019532
     }
   ],
   "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": 0.08977657719512422,
-    "producer_p95_seconds": 0.22217336600078852,
-    "gpu_update_seconds": 0.9569608036877194,
-    "producer_batches_per_second": 11.138762818129695,
-    "gpu_batches_per_second": 1.0449748789568245,
-    "producer_headroom": 10.65935941852428,
+    "producer_batch_seconds": 0.08805437375787051,
+    "producer_p95_seconds": 0.2127668730026926,
+    "gpu_update_seconds": 0.952792036810783,
+    "producer_batches_per_second": 11.356619294684583,
+    "gpu_batches_per_second": 1.0495469749592294,
+    "producer_headroom": 10.82049642906716,
     "sustainable": true,
-    "buffer_seconds": 60.28853063232632,
+    "buffer_seconds": 60.025898319079324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312662.31897651224,
-    "upload_bytes_per_second": 4776004.706385802,
+    "generation_bytes_per_second": 297998.5470893914,
+    "upload_bytes_per_second": 4787096.310507735,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.09005524608159,
+    "estimated_full_shards_per_hour": 63.94355115424448,
     "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": 0.9283436340010667,
-      "bytes": 2568999,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.7634025150000525,
       "bytes": 6258608,
@@ -562,6 +556,12 @@
       "bytes": 11874470,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.439283288000297,
+      "bytes": 6018491,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 944400, "loss": 0.5279709100723267, "lr": 3e-05, "grad_norm": 0.32609924167972903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138233576903, "tokens_per_second": 144423.03887124828, "data_wait_s": 1.3274002412799746e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944500, "loss": 0.577720582485199, "lr": 3e-05, "grad_norm": 0.35585286234893304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138248100173, "tokens_per_second": 150613.3662130495, "data_wait_s": 1.7441001546103507e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944600, "loss": 0.555435836315155, "lr": 3e-05, "grad_norm": 0.35039679697845094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138262793802, "tokens_per_second": 153374.08155014302, "data_wait_s": 1.309999788645655e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 944700, "loss": 0.5725507736206055, "lr": 3e-05, "grad_norm": 0.34738436125650035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138277526113, "tokens_per_second": 152562.25768287326, "data_wait_s": 1.4356002793647349e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:965ed443d95804a62bd5001dd96be135c7857d0e8f57b08e05e4ee2ef5199254
-size 417642
+oid sha256:55b22316dc292ec2b0867650d0f83e73c0906a1c7491f66852dedd59a713e880
+size 419760