Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76293.79164936279,
-      "required_pages_per_second_upper_estimate": 0.006612952495319659
+      "required_token_positions_per_second": 76240.36309572149,
+      "required_pages_per_second_upper_estimate": 0.006608321443703462
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9536.723956170348,
-      "required_pages_per_second_upper_estimate": 0.0038043417728459983
+      "required_token_positions_per_second": 9530.045386965186,
+      "required_pages_per_second_upper_estimate": 0.003801677591736551
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19073.447912340696,
-      "required_pages_per_second_upper_estimate": 0.0044123064933017
+      "required_token_positions_per_second": 19060.090773930373,
+      "required_pages_per_second_upper_estimate": 0.004409216554408079
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28610.171868511046,
-      "required_pages_per_second_upper_estimate": 0.0017422093608213294
+      "required_token_positions_per_second": 28590.13616089556,
+      "required_pages_per_second_upper_estimate": 0.00174098929134678
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28610.171868511046,
-      "required_pages_per_second_upper_estimate": 0.0022024902284626113
+      "required_token_positions_per_second": 28590.13616089556,
+      "required_pages_per_second_upper_estimate": 0.0022009478242279827
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28610.171868511046,
-      "required_pages_per_second_upper_estimate": 0.35524258252121443
+      "required_token_positions_per_second": 28590.13616089556,
+      "required_pages_per_second_upper_estimate": 0.35499380608782993
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04707358759374891,
-    "producer_p95_seconds": 0.1140000240000063,
-    "gpu_update_seconds": 0.9663695879586787,
-    "producer_batches_per_second": 21.243335193190035,
-    "gpu_batches_per_second": 1.0348007764941785,
-    "producer_headroom": 20.52891307751115,
+    "producer_batch_seconds": 0.05286207063282333,
+    "producer_p95_seconds": 0.11887485300007938,
+    "gpu_update_seconds": 0.9670468109842661,
+    "producer_batches_per_second": 18.917155306797156,
+    "gpu_batches_per_second": 1.0340761053564655,
+    "producer_headroom": 18.293774712332276,
     "sustainable": true,
-    "buffer_seconds": 60.88128404139675,
+    "buffer_seconds": 60.92394909200877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 58,
+    "upload_pages_per_commit": 53,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 40300, "loss": 0.6501443609595299, "lr": 0.0003, "grad_norm": 0.14097667619387938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5897110808, "tokens_per_second": 150874.0284524783, "data_wait_s": 9.098999953494058e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 40400, "loss": 0.6147063747048378, "lr": 0.0003, "grad_norm": 0.12158116090798718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5911786558, "tokens_per_second": 151573.0581806831, "data_wait_s": 8.046999937505461e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 40500, "loss": 0.6719138324260712, "lr": 0.0003, "grad_norm": 0.14287059746612243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5926359613, "tokens_per_second": 150518.8209185117, "data_wait_s": 1.3272999922264717e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 40600, "loss": 0.6330451220273972, "lr": 0.0003, "grad_norm": 0.12777892232787574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5941002069, "tokens_per_second": 151153.97869570347, "data_wait_s": 1.0009000106947497e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d4a66e87d914fb69e79cef4e9f155f632726dded77ee41f3a87c7bc621630bc
-size 10755
+oid sha256:8171df520dd02a4473276b0d2e771d7aeedfcbb9ce149cf38aa55cf8ee1a9979
+size 12873