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": 76903.07269691234,
-      "required_pages_per_second_upper_estimate": 0.005320333008885247
+      "required_token_positions_per_second": 75717.37428319365,
+      "required_pages_per_second_upper_estimate": 0.005238303641424828
     },
     {
       "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": 9612.884087114042,
-      "required_pages_per_second_upper_estimate": 0.004024713827428021
+      "required_token_positions_per_second": 9464.671785399207,
+      "required_pages_per_second_upper_estimate": 0.003962660431724827
     },
     {
       "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": 19225.768174228084,
-      "required_pages_per_second_upper_estimate": 0.004004220493089498
+      "required_token_positions_per_second": 18929.343570798414,
+      "required_pages_per_second_upper_estimate": 0.003942483065437577
     },
     {
       "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": 28838.652261342122,
-      "required_pages_per_second_upper_estimate": 0.0017288331689951125
+      "required_token_positions_per_second": 28394.015356197622,
+      "required_pages_per_second_upper_estimate": 0.0017021778654529433
     },
     {
       "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": 28838.652261342122,
-      "required_pages_per_second_upper_estimate": 0.002242078365493744
+      "required_token_positions_per_second": 28394.015356197622,
+      "required_pages_per_second_upper_estimate": 0.0022075098018698143
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28838.652261342122,
-      "required_pages_per_second_upper_estimate": 0.3699112666761858
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28394.015356197622,
+      "required_pages_per_second_upper_estimate": 0.364207941870905
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09804384417972756,
-    "producer_p95_seconds": 0.2815186830011953,
-    "gpu_update_seconds": 0.9587133181345586,
-    "producer_batches_per_second": 10.19951847427428,
-    "gpu_batches_per_second": 1.0430646795913674,
-    "producer_headroom": 9.778414199846225,
+    "producer_batch_seconds": 0.22838714199997412,
+    "producer_p95_seconds": 0.28193746699980693,
+    "gpu_update_seconds": 0.9737263170834065,
+    "producer_batches_per_second": 4.378530206398893,
+    "gpu_batches_per_second": 1.0269826156032127,
+    "producer_headroom": 4.263490092115242,
     "sustainable": true,
-    "buffer_seconds": 60.398939042477195,
+    "buffer_seconds": 60.3710316591712,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 528194.730934447,
-    "upload_bytes_per_second": 5016502.409756339,
+    "generation_bytes_per_second": 541021.5621002669,
+    "upload_bytes_per_second": 5041119.1567016505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.33829351449069,
+    "estimated_full_shards_per_hour": 116.09063288932806,
     "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": 0.8717172680007934,
-      "bytes": 283568,
-      "pages": 8,
-      "shards": 1
-    },
     {
       "seconds": 14.44776459899731,
       "bytes": 11619395,
@@ -562,6 +556,12 @@
       "bytes": 6135465,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.8637312579994614,
+      "bytes": 2324435,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -256,3 +256,4 @@
 {"step": 422700, "loss": 0.5855945348739624, "lr": 0.0003, "grad_norm": 0.15746786691532302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61871966723, "tokens_per_second": 150422.87907787634, "data_wait_s": 1.0731997463153675e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 422800, "loss": 0.5643939971923828, "lr": 0.0003, "grad_norm": 0.18316346773762682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61886690249, "tokens_per_second": 152335.3523300517, "data_wait_s": 9.429000783711672e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 422900, "loss": 0.5900300741195679, "lr": 0.0003, "grad_norm": 0.16598900074444947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61901186327, "tokens_per_second": 150067.61559428167, "data_wait_s": 1.1060998076573014e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 423000, "loss": 0.5987367033958435, "lr": 0.0003, "grad_norm": 0.1648033516490014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61915632500, "tokens_per_second": 150065.99089957593, "data_wait_s": 1.168100061477162e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:430f8452c7eae8f4037f8fb731fc8e79cdfd3bf7278c8c2bad646da52e93e7ba
-size 550524
+oid sha256:292949b452ec9b8b36905b5b7e2aba5d7bad1c7fb3042a89c9b3b4aa2fde9817
+size 552642