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": 77040.99085543091,
-      "required_pages_per_second_upper_estimate": 0.005329874507105779
+      "required_token_positions_per_second": 76446.76820649613,
+      "required_pages_per_second_upper_estimate": 0.005288764805465965
     },
     {
       "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": 9630.123856928863,
-      "required_pages_per_second_upper_estimate": 0.004031931759042155
+      "required_token_positions_per_second": 9555.846025812016,
+      "required_pages_per_second_upper_estimate": 0.004000833182251026
     },
     {
       "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": 19260.247713857727,
-      "required_pages_per_second_upper_estimate": 0.004011401671907746
+      "required_token_positions_per_second": 19111.69205162403,
+      "required_pages_per_second_upper_estimate": 0.003980461445140733
     },
     {
       "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": 28890.371570786592,
-      "required_pages_per_second_upper_estimate": 0.0017319336626254957
+      "required_token_positions_per_second": 28667.538077436046,
+      "required_pages_per_second_upper_estimate": 0.0017185751349461738
     },
     {
       "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": 28890.371570786592,
-      "required_pages_per_second_upper_estimate": 0.002246099314313851
+      "required_token_positions_per_second": 28667.538077436046,
+      "required_pages_per_second_upper_estimate": 0.0022287749903467987
     },
     {
       "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": 28890.371570786592,
-      "required_pages_per_second_upper_estimate": 0.37057466644587156
+      "required_token_positions_per_second": 28667.538077436046,
+      "required_pages_per_second_upper_estimate": 0.36771639765313485
     }
   ],
   "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": 0.08575899957054389,
-    "producer_p95_seconds": 0.30200607200094964,
-    "gpu_update_seconds": 0.9569970373090371,
-    "producer_batches_per_second": 11.660583787214273,
-    "gpu_batches_per_second": 1.0449353143368993,
-    "producer_headroom": 11.15914413765785,
+    "producer_batch_seconds": 0.3960316080233497,
+    "producer_p95_seconds": 0.3223909939988516,
+    "gpu_update_seconds": 0.9644357993113293,
+    "producer_batches_per_second": 2.525050980125407,
+    "gpu_batches_per_second": 1.0368756538424495,
+    "producer_headroom": 2.435249560319102,
     "sustainable": true,
-    "buffer_seconds": 60.290813350469335,
+    "buffer_seconds": 60.75945535661374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348567.8030750391,
-    "upload_bytes_per_second": 5509888.262520985,
+    "generation_bytes_per_second": 320203.31389787275,
+    "upload_bytes_per_second": 5582551.380920806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.79453629673367,
+    "estimated_full_shards_per_hour": 68.70817721082818,
     "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": 0.8838252950008609,
-      "bytes": 2597092,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0776322439996875,
       "bytes": 6282866,
@@ -562,6 +556,12 @@
       "bytes": 7649877,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2835066039988305,
+      "bytes": 10437972,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 520600, "loss": 0.5928990244865417, "lr": 0.0003, "grad_norm": 0.16258489342446983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76198092963, "tokens_per_second": 152360.54021783645, "data_wait_s": 1.4529003237839788e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520700, "loss": 0.5461702346801758, "lr": 0.0003, "grad_norm": 0.15240206710258022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76212806497, "tokens_per_second": 153434.50988860332, "data_wait_s": 1.4102995919529349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520800, "loss": 0.560724675655365, "lr": 0.0003, "grad_norm": 0.14476038762202956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76227378556, "tokens_per_second": 151738.8896835203, "data_wait_s": 1.322099706158042e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 520900, "loss": 0.5985284447669983, "lr": 0.0003, "grad_norm": 0.14101838737853054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76242166162, "tokens_per_second": 154491.59539069392, "data_wait_s": 1.4563003787770867e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef9776c7d7e3f4b4a6b748065e0bbfcac5b509e45ef14422bbfc95025519f02b
-size 763890
+oid sha256:1d2d9b6b63b8a32550ea49b4842a910d55c79d2b734573415c0354b84c90dcbf
+size 766008