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": 75658.47014854083,
-      "required_pages_per_second_upper_estimate": 0.006557884437836034
+      "required_token_positions_per_second": 75723.0115804251,
+      "required_pages_per_second_upper_estimate": 0.006563478725572999
     },
     {
       "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": 9457.308768567604,
-      "required_pages_per_second_upper_estimate": 0.003772661867148398
+      "required_token_positions_per_second": 9465.376447553137,
+      "required_pages_per_second_upper_estimate": 0.003775880184918277
     },
     {
       "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": 18914.617537135207,
-      "required_pages_per_second_upper_estimate": 0.004375563881317042
+      "required_token_positions_per_second": 18930.752895106274,
+      "required_pages_per_second_upper_estimate": 0.004379296512411046
     },
     {
       "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": 28371.926305702807,
-      "required_pages_per_second_upper_estimate": 0.0017277014560241723
+      "required_token_positions_per_second": 28396.129342659413,
+      "required_pages_per_second_upper_estimate": 0.0017291752939913084
     },
     {
       "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": 28371.926305702807,
-      "required_pages_per_second_upper_estimate": 0.002184149425531705
+      "required_token_positions_per_second": 28396.129342659413,
+      "required_pages_per_second_upper_estimate": 0.002186012642314919
     },
     {
       "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": 28371.926305702807,
-      "required_pages_per_second_upper_estimate": 0.3522843699877424
+      "required_token_positions_per_second": 28396.129342659413,
+      "required_pages_per_second_upper_estimate": 0.3525848907043895
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.39677208125010566,
-    "producer_p95_seconds": 0.3930985309998505,
-    "gpu_update_seconds": 0.9744844147026669,
-    "producer_batches_per_second": 2.520338620724801,
-    "gpu_batches_per_second": 1.0261836771449222,
-    "producer_headroom": 2.4560307056695345,
+    "producer_batch_seconds": 0.15124620003138034,
+    "producer_p95_seconds": 0.37854810400313,
+    "gpu_update_seconds": 0.973653826772246,
+    "producer_batches_per_second": 6.611736359607854,
+    "gpu_batches_per_second": 1.0270590763404013,
+    "producer_headroom": 6.437542408141386,
     "sustainable": true,
-    "buffer_seconds": 60.41803371156535,
+    "buffer_seconds": 60.36653725987925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426179.0858958272,
-    "upload_bytes_per_second": 3745452.8459881484,
+    "generation_bytes_per_second": 406386.5938844186,
+    "upload_bytes_per_second": 3781637.11005544,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.44811089187729,
+    "estimated_full_shards_per_hour": 87.20110285186213,
     "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": 1.245932874000573,
-      "bytes": 7158906,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 0.8130094469997857,
       "bytes": 5715102,
@@ -562,6 +556,12 @@
       "bytes": 1243052,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.3106369909946807,
+      "bytes": 11452061,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -377,3 +377,4 @@
 {"step": 37700, "loss": 0.6304601151496172, "lr": 0.0003, "grad_norm": 0.1387206411435085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5516753921, "tokens_per_second": 148644.81277852427, "data_wait_s": 1.16259980131872e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37800, "loss": 0.7046154029667377, "lr": 0.0003, "grad_norm": 0.155621582148506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5531510173, "tokens_per_second": 149407.14831093606, "data_wait_s": 1.4583994925487787e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37900, "loss": 0.6656279861927032, "lr": 0.0003, "grad_norm": 0.14468059415255227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5546372380, "tokens_per_second": 151590.1060227946, "data_wait_s": 1.3238000974524766e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 38000, "loss": 0.6564249563962221, "lr": 0.0003, "grad_norm": 0.13602110274995644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5560882393, "tokens_per_second": 147321.59861726186, "data_wait_s": 1.0757001291494817e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc9dad9f407204cf77466d13ee9d1363b85c13ef4c50c40a91e758e2f13c54f6
-size 802594
+oid sha256:c159b1305b9b0574263c6c1ba74b4ede42683ff31640f31f1b38e7e929d2234a
+size 804712