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": 76410.98879910566,
-      "required_pages_per_second_upper_estimate": 0.00528628950304303
+      "required_token_positions_per_second": 76559.42827113728,
+      "required_pages_per_second_upper_estimate": 0.005296558890145206
     },
     {
       "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": 9551.373599888208,
-      "required_pages_per_second_upper_estimate": 0.003998960670911602
+      "required_token_positions_per_second": 9569.92853389216,
+      "required_pages_per_second_upper_estimate": 0.004006729234307974
     },
     {
       "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": 19102.747199776415,
-      "required_pages_per_second_upper_estimate": 0.003978598468392481
+      "required_token_positions_per_second": 19139.85706778432,
+      "required_pages_per_second_upper_estimate": 0.003986327475245496
     },
     {
       "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": 28654.120799664623,
-      "required_pages_per_second_upper_estimate": 0.001717770789630774
+      "required_token_positions_per_second": 28709.78560167648,
+      "required_pages_per_second_upper_estimate": 0.0017211078095161535
     },
     {
       "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": 28654.120799664623,
-      "required_pages_per_second_upper_estimate": 0.0022277318560164384
+      "required_token_positions_per_second": 28709.78560167648,
+      "required_pages_per_second_upper_estimate": 0.002232059549529272
     },
     {
       "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": 28654.120799664623,
-      "required_pages_per_second_upper_estimate": 0.36754429522023346
+      "required_token_positions_per_second": 28709.78560167648,
+      "required_pages_per_second_upper_estimate": 0.3682583035322338
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11104079755486396,
-    "producer_p95_seconds": 0.23255385700031184,
-    "gpu_update_seconds": 0.9648873958932846,
-    "producer_batches_per_second": 9.005699004511488,
-    "gpu_batches_per_second": 1.0363903645712031,
-    "producer_headroom": 8.689485460661835,
+    "producer_batch_seconds": 0.08143887151581453,
+    "producer_p95_seconds": 0.23460413000066183,
+    "gpu_update_seconds": 0.9630165959297697,
+    "producer_batches_per_second": 12.279148536651949,
+    "gpu_batches_per_second": 1.0384037037643403,
+    "producer_headroom": 11.825023824682573,
     "sustainable": true,
-    "buffer_seconds": 60.78790594127693,
+    "buffer_seconds": 60.67004554357549,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285666.7520374398,
-    "upload_bytes_per_second": 5037474.848808913,
+    "generation_bytes_per_second": 297927.88501695596,
+    "upload_bytes_per_second": 5035894.375823741,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.2974350055923,
+    "estimated_full_shards_per_hour": 63.9283887184287,
     "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": 1.1540112700004101,
-      "bytes": 11807839,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 0.8518780800004606,
       "bytes": 2360936,
@@ -562,6 +556,12 @@
       "bytes": 834051,
       "pages": 23,
       "shards": 1
+    },
+    {
+      "seconds": 1.34710112600078,
+      "bytes": 12646439,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -201,3 +201,4 @@
 {"step": 460300, "loss": 0.6818230748176575, "lr": 0.0003, "grad_norm": 0.1774673164541843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67373822275, "tokens_per_second": 152216.01214944094, "data_wait_s": 1.3736000255448744e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460400, "loss": 0.546209990978241, "lr": 0.0003, "grad_norm": 0.15693226266937596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67388442938, "tokens_per_second": 152288.8353781708, "data_wait_s": 1.4305998774943873e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460500, "loss": 0.5352906584739685, "lr": 0.0003, "grad_norm": 0.1556287384075062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67403073268, "tokens_per_second": 152620.37214929378, "data_wait_s": 1.1044998245779425e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 460600, "loss": 0.563446581363678, "lr": 0.0003, "grad_norm": 0.16265674861013651, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67417736660, "tokens_per_second": 152387.3588634003, "data_wait_s": 1.5427001926582307e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:745fef9fa40dadd60194ad5addad7871c4e393d018ec2d86030620a76d5ff892
-size 433251
+oid sha256:6866048e2b996ec6f70996255ec262d7fe6c78aa45f857a3ce7f6be53f86f2b4
+size 435369