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": 76410.74403823842,
-      "required_pages_per_second_upper_estimate": 0.005286272569918311
+      "required_token_positions_per_second": 76981.82657808445,
+      "required_pages_per_second_upper_estimate": 0.005325781385118914
     },
     {
       "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": 9551.343004779803,
-      "required_pages_per_second_upper_estimate": 0.003998947861378611
+      "required_token_positions_per_second": 9622.728322260557,
+      "required_pages_per_second_upper_estimate": 0.004028835403112861
     },
     {
       "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": 19102.686009559606,
-      "required_pages_per_second_upper_estimate": 0.003978585724084014
+      "required_token_positions_per_second": 19245.456644521113,
+      "required_pages_per_second_upper_estimate": 0.004008321082231659
     },
     {
       "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": 28654.029014339412,
-      "required_pages_per_second_upper_estimate": 0.0017177652872406757
+      "required_token_positions_per_second": 28868.18496678167,
+      "required_pages_per_second_upper_estimate": 0.0017306036095923833
     },
     {
       "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": 28654.029014339412,
-      "required_pages_per_second_upper_estimate": 0.0022277247201111243
+      "required_token_positions_per_second": 28868.18496678167,
+      "required_pages_per_second_upper_estimate": 0.0022443744034410257
     },
     {
       "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": 28654.029014339412,
-      "required_pages_per_second_upper_estimate": 0.3675431178966331
+      "required_token_positions_per_second": 28868.18496678167,
+      "required_pages_per_second_upper_estimate": 0.3702900805118158
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.45219762534370034,
-    "producer_p95_seconds": 0.1948362620005355,
-    "gpu_update_seconds": 0.9648904866454919,
-    "producer_batches_per_second": 2.211422493074645,
-    "gpu_batches_per_second": 1.0363870447894752,
-    "producer_headroom": 2.133780525521581,
+    "producer_batch_seconds": 0.09387903632799066,
+    "producer_p95_seconds": 0.18108731299980718,
+    "gpu_update_seconds": 0.9577325360709125,
+    "producer_batches_per_second": 10.652005379627479,
+    "gpu_batches_per_second": 1.0441328474675085,
+    "producer_headroom": 10.201772126471628,
     "sustainable": true,
-    "buffer_seconds": 60.78810065866599,
+    "buffer_seconds": 60.337149772467484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263523.4123833888,
-    "upload_bytes_per_second": 4554930.553710419,
+    "generation_bytes_per_second": 265069.2454266402,
+    "upload_bytes_per_second": 4563708.842623039,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.54598978639839,
+    "estimated_full_shards_per_hour": 56.87768957232861,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 7840439,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.00542208799925,
+      "bytes": 5360697,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 534000, "loss": 0.618455171585083, "lr": 0.0003, "grad_norm": 0.17522603426261435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78160528918, "tokens_per_second": 152357.51023718226, "data_wait_s": 7.5360003393143415e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 534100, "loss": 0.6180509924888611, "lr": 0.0003, "grad_norm": 0.15880729440507382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78175250848, "tokens_per_second": 153266.85742901676, "data_wait_s": 1.0710999049479142e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 534200, "loss": 0.545046329498291, "lr": 0.0003, "grad_norm": 0.14561254477901242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78190043077, "tokens_per_second": 154572.469506171, "data_wait_s": 1.1749998520826921e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 534300, "loss": 0.5961855053901672, "lr": 0.0003, "grad_norm": 0.15559007899240454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78204758922, "tokens_per_second": 153818.65808628243, "data_wait_s": 1.0103998647537082e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:766fc78bcd3538a2bf1169c8b4899fdc083c4ee0e0de419a0e3cba0a5e3b7868
-size 195804
+oid sha256:c92667cbc40865890031f8f3f3eb136a36ff9fa6f830f83c90a1583e3467fa08
+size 197922