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": 74482.33602151592,
-      "required_pages_per_second_upper_estimate": 0.0064559401125902065
+      "required_token_positions_per_second": 74310.83304224495,
+      "required_pages_per_second_upper_estimate": 0.0064410746690173265
     },
     {
       "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": 9310.29200268949,
-      "required_pages_per_second_upper_estimate": 0.0037140146811430867
+      "required_token_positions_per_second": 9288.854130280619,
+      "required_pages_per_second_upper_estimate": 0.0037054627933144325
     },
     {
       "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": 18620.58400537898,
-      "required_pages_per_second_upper_estimate": 0.004307544398558656
+      "required_token_positions_per_second": 18577.708260561238,
+      "required_pages_per_second_upper_estimate": 0.004297625849582407
     },
     {
       "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": 27930.87600806847,
-      "required_pages_per_second_upper_estimate": 0.0017008438069103164
+      "required_token_positions_per_second": 27866.562390841857,
+      "required_pages_per_second_upper_estimate": 0.0016969274450486876
     },
     {
       "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": 27930.87600806847,
-      "required_pages_per_second_upper_estimate": 0.0021501961527144488
+      "required_token_positions_per_second": 27866.562390841857,
+      "required_pages_per_second_upper_estimate": 0.002145245112428857
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27930.87600806847,
-      "required_pages_per_second_upper_estimate": 0.3468080013915153
+      "required_token_positions_per_second": 27866.562390841857,
+      "required_pages_per_second_upper_estimate": 0.3460094415093914
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13751513792973924,
-    "producer_p95_seconds": 0.3182933470015996,
-    "gpu_update_seconds": 0.9898722829893787,
-    "producer_batches_per_second": 7.271926676981054,
-    "gpu_batches_per_second": 1.0102313370973839,
-    "producer_headroom": 7.198278661474602,
+    "producer_batch_seconds": 0.08701305005470772,
+    "producer_p95_seconds": 0.24190976200043224,
+    "gpu_update_seconds": 0.9921568226544626,
+    "producer_batches_per_second": 11.492528986988388,
+    "gpu_batches_per_second": 1.0079051790669074,
+    "producer_headroom": 11.402391043994708,
     "sustainable": true,
-    "buffer_seconds": 60.3822092623521,
+    "buffer_seconds": 60.52156618192222,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368321.3692730293,
-    "upload_bytes_per_second": 5529517.118044717,
+    "generation_bytes_per_second": 383961.4247300378,
+    "upload_bytes_per_second": 5464409.796599209,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.0331917633358,
+    "estimated_full_shards_per_hour": 82.38918358255243,
     "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": 0.940834745999382,
-      "bytes": 6400703,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1989260670015938,
       "bytes": 2439937,
@@ -562,6 +556,12 @@
       "bytes": 11962174,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.8873195009982737,
+      "bytes": 1168035,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -208,3 +208,4 @@
 {"step": 335800, "loss": 0.603604007512331, "lr": 0.0003, "grad_norm": 0.16256666828737928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49152367835, "tokens_per_second": 143218.72598825878, "data_wait_s": 1.428500036126934e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 335900, "loss": 0.667185828089714, "lr": 0.0003, "grad_norm": 0.19191534515916744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49167108350, "tokens_per_second": 149633.01696525273, "data_wait_s": 1.2935000995639712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 336000, "loss": 0.6260675601661205, "lr": 0.0003, "grad_norm": 0.15267251646496335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49181851800, "tokens_per_second": 149769.7196090554, "data_wait_s": 8.497001545038074e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 336100, "loss": 0.6162252761423588, "lr": 0.0003, "grad_norm": 0.15455633509498237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49196560391, "tokens_per_second": 149375.19740666906, "data_wait_s": 1.2152999261161312e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08c14556df7cba0e516cf04348e74b320e866e98f17e66e546ece456e88e3156
-size 448077
+oid sha256:9a5e44cd9ef381664c486f01fc0ab465e581b3b67b97167e3c28e8a505fcb969
+size 450195