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": 77231.26058591035,
-      "required_pages_per_second_upper_estimate": 0.0053430378085469465
+      "required_token_positions_per_second": 77081.05680434106,
+      "required_pages_per_second_upper_estimate": 0.0053326463624171925
     },
     {
       "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": 9653.907573238794,
-      "required_pages_per_second_upper_estimate": 0.00404188950440065
+      "required_token_positions_per_second": 9635.132100542633,
+      "required_pages_per_second_upper_estimate": 0.004034028606059221
     },
     {
       "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": 19307.815146477587,
-      "required_pages_per_second_upper_estimate": 0.0040213087136849075
+      "required_token_positions_per_second": 19270.264201085265,
+      "required_pages_per_second_upper_estimate": 0.0040134878420447105
     },
     {
       "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": 28961.722719716377,
-      "required_pages_per_second_upper_estimate": 0.00173621105505694
+      "required_token_positions_per_second": 28905.396301627898,
+      "required_pages_per_second_upper_estimate": 0.0017328343722979962
     },
     {
       "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": 28961.722719716377,
-      "required_pages_per_second_upper_estimate": 0.002251646552302605
+      "required_token_positions_per_second": 28905.396301627898,
+      "required_pages_per_second_upper_estimate": 0.0022472674210499556
     },
     {
       "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": 28961.722719716377,
-      "required_pages_per_second_upper_estimate": 0.371489882373448
+      "required_token_positions_per_second": 28905.396301627898,
+      "required_pages_per_second_upper_estimate": 0.37076738756080474
     }
   ],
   "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": 0.14833737332065766,
-    "producer_p95_seconds": 0.35492088200408034,
-    "gpu_update_seconds": 0.954639344750648,
-    "producer_batches_per_second": 6.741389426104518,
-    "gpu_batches_per_second": 1.047516012721223,
-    "producer_headroom": 6.435595584445364,
+    "producer_batch_seconds": 0.5010926698827802,
+    "producer_p95_seconds": 0.35023676700075157,
+    "gpu_update_seconds": 0.9564995999879412,
+    "producer_batches_per_second": 1.9956388510610787,
+    "gpu_batches_per_second": 1.0454787435484627,
+    "producer_headroom": 1.9088277627603165,
     "sustainable": true,
-    "buffer_seconds": 60.14227871929083,
+    "buffer_seconds": 60.259474799240294,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352768.89833070716,
-    "upload_bytes_per_second": 4547559.729268987,
+    "generation_bytes_per_second": 307514.50333523325,
+    "upload_bytes_per_second": 4551507.707987857,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.69599354210769,
+    "estimated_full_shards_per_hour": 65.98545384447812,
     "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": 1.4696884980003233,
-      "bytes": 5989675,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.5069389330019476,
       "bytes": 6220019,
@@ -562,6 +556,12 @@
       "bytes": 7219452,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1526426790005644,
+      "bytes": 4920482,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -406,3 +406,4 @@
 {"step": 800600, "loss": 0.5685527324676514, "lr": 3e-05, "grad_norm": 0.2735512879455177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117184702810, "tokens_per_second": 152434.87473564062, "data_wait_s": 1.0414005373604596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800700, "loss": 0.5554490089416504, "lr": 3e-05, "grad_norm": 0.2789289776740081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117199349780, "tokens_per_second": 152652.23096207983, "data_wait_s": 1.6323996533174068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800800, "loss": 0.5392630696296692, "lr": 3e-05, "grad_norm": 0.2653088635396031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117213898762, "tokens_per_second": 141825.97205698508, "data_wait_s": 1.095700281439349e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 800900, "loss": 0.548346757888794, "lr": 3e-05, "grad_norm": 0.2749633752369359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117228365568, "tokens_per_second": 149780.98371397235, "data_wait_s": 6.10799907008186e-06, "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 sha256:8d2a2801a0d1fb70f138c8beaeb407809b7005954993c4430b7e03ac3187f918
-size 870573
+oid sha256:27040fbbf29b34dd4a6a92517a79de309098dceea6648777f884b40722032d6f
+size 872691