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": 76540.01055035356,
-      "required_pages_per_second_upper_estimate": 0.0052952155271660676
+      "required_token_positions_per_second": 76930.48427952675,
+      "required_pages_per_second_upper_estimate": 0.005322229405774146
     },
     {
       "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": 9567.501318794195,
-      "required_pages_per_second_upper_estimate": 0.004005713010032471
+      "required_token_positions_per_second": 9616.310534940843,
+      "required_pages_per_second_upper_estimate": 0.004026148409580736
     },
     {
       "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": 19135.00263758839,
-      "required_pages_per_second_upper_estimate": 0.003985316425455617
+      "required_token_positions_per_second": 19232.621069881687,
+      "required_pages_per_second_upper_estimate": 0.004005647770531132
     },
     {
       "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": 28702.503956382585,
-      "required_pages_per_second_upper_estimate": 0.001720671285999213
+      "required_token_positions_per_second": 28848.93160482253,
+      "required_pages_per_second_upper_estimate": 0.0017294493999411144
     },
     {
       "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": 28702.503956382585,
-      "required_pages_per_second_upper_estimate": 0.0022314934336362447
+      "required_token_positions_per_second": 28848.93160482253,
+      "required_pages_per_second_upper_estimate": 0.0022428775392352917
     },
     {
       "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": 28702.503956382585,
-      "required_pages_per_second_upper_estimate": 0.3681649024048253
+      "required_token_positions_per_second": 28848.93160482253,
+      "required_pages_per_second_upper_estimate": 0.37004311905725334
     }
   ],
   "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": 0.10122918901559785,
-    "producer_p95_seconds": 0.19187767899893515,
-    "gpu_update_seconds": 0.9632609072022061,
-    "producer_batches_per_second": 9.878573657701786,
-    "gpu_batches_per_second": 1.0381403340705506,
-    "producer_headroom": 9.51564382338164,
+    "producer_batch_seconds": 0.07993122249997953,
+    "producer_p95_seconds": 0.1883832599996822,
+    "gpu_update_seconds": 0.958371712988436,
+    "producer_batches_per_second": 12.510755731282055,
+    "gpu_batches_per_second": 1.043436472975352,
+    "producer_headroom": 11.989954400968676,
     "sustainable": true,
-    "buffer_seconds": 60.68543715373899,
+    "buffer_seconds": 60.37741791827147,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335843.0123017782,
-    "upload_bytes_per_second": 5020987.034443281,
+    "generation_bytes_per_second": 336134.81550384755,
+    "upload_bytes_per_second": 5020597.0200018175,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.06409241476068,
+    "estimated_full_shards_per_hour": 72.12670658909387,
     "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": 1.037127201000203,
-      "bytes": 7612192,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9605875500001275,
       "bytes": 5330342,
@@ -562,6 +556,12 @@
       "bytes": 6831941,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.0320983010005875,
+      "bytes": 7553870,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -97,3 +97,4 @@
 {"step": 406800, "loss": 0.546454906463623, "lr": 0.0003, "grad_norm": 0.18739730907034083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59544924867, "tokens_per_second": 110664.96534534721, "data_wait_s": 1.964900002349168e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406900, "loss": 0.5444069504737854, "lr": 0.0003, "grad_norm": 0.1747523611226791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59559556766, "tokens_per_second": 152496.75751867174, "data_wait_s": 8.297998647321947e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 407000, "loss": 0.572537362575531, "lr": 0.0003, "grad_norm": 0.13790204543994558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59574217989, "tokens_per_second": 152536.78562929915, "data_wait_s": 1.2856000466854312e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 407100, "loss": 0.5529351830482483, "lr": 0.0003, "grad_norm": 0.16667881099229423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59588873855, "tokens_per_second": 146776.5069391976, "data_wait_s": 8.43499947222881e-06, "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:a2d157e67a3f41a780e408d817e148629a6baef014020db0a1518af21136bafe
-size 211413
+oid sha256:3b9ec5d20beb93315dea22a2f90a7947590c352c79684972f7dcdef242bbe680
+size 213531