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": 76939.71339558266,
-      "required_pages_per_second_upper_estimate": 0.005322867897437394
+      "required_token_positions_per_second": 76109.12156782419,
+      "required_pages_per_second_upper_estimate": 0.0052654056275545005
     },
     {
       "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": 9617.464174447832,
-      "required_pages_per_second_upper_estimate": 0.00402663141435158
+      "required_token_positions_per_second": 9513.640195978023,
+      "required_pages_per_second_upper_estimate": 0.003983162482657484
     },
     {
       "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": 19234.928348895664,
-      "required_pages_per_second_upper_estimate": 0.0040061283159027045
+      "required_token_positions_per_second": 19027.280391956047,
+      "required_pages_per_second_upper_estimate": 0.003962880722517055
     },
     {
       "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": 28852.392523343497,
-      "required_pages_per_second_upper_estimate": 0.0017296568767219299
+      "required_token_positions_per_second": 28540.92058793407,
+      "required_pages_per_second_upper_estimate": 0.0017109846097842424
     },
     {
       "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": 28852.392523343497,
-      "required_pages_per_second_upper_estimate": 0.002243146610427325
+      "required_token_positions_per_second": 28540.92058793407,
+      "required_pages_per_second_upper_estimate": 0.002218931037469498
     },
     {
       "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": 28852.392523343497,
-      "required_pages_per_second_upper_estimate": 0.37008751200399553
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28540.92058793407,
+      "required_pages_per_second_upper_estimate": 0.3660922844490716
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09945462215608813,
-    "producer_p95_seconds": 0.36299237400089623,
-    "gpu_update_seconds": 0.9582567538421966,
-    "producer_batches_per_second": 10.054836852434663,
-    "gpu_batches_per_second": 1.0435616508732457,
-    "producer_headroom": 9.63511532262693,
+    "producer_batch_seconds": 0.27881982812516526,
+    "producer_p95_seconds": 0.373849686999165,
+    "gpu_update_seconds": 0.9687143732738754,
+    "producer_batches_per_second": 3.5865455004551867,
+    "gpu_batches_per_second": 1.032296028209421,
+    "producer_headroom": 3.4743381766916843,
     "sustainable": true,
-    "buffer_seconds": 60.37017549205838,
+    "buffer_seconds": 60.060291142980276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 545446.5974246027,
-    "upload_bytes_per_second": 5032827.53437783,
+    "generation_bytes_per_second": 512106.81724401616,
+    "upload_bytes_per_second": 5045484.191235549,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.04014246038018,
+    "estimated_full_shards_per_hour": 109.88620174398768,
     "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": 1.0934254949970637,
-      "bytes": 6083579,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.342605496996839,
       "bytes": 1094629,
@@ -562,6 +556,12 @@
       "bytes": 5803988,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.751644532996579,
+      "bytes": 10468749,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -437,3 +437,4 @@
 {"step": 968700, "loss": 0.5292211771011353, "lr": 3e-05, "grad_norm": 0.35714355296152034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141791288879, "tokens_per_second": 150314.5316953439, "data_wait_s": 8.096001693047583e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 968800, "loss": 0.5146040320396423, "lr": 3e-05, "grad_norm": 0.35547420129965146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141805994106, "tokens_per_second": 153702.53724853892, "data_wait_s": 1.4299999747890979e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 968900, "loss": 0.5428612232208252, "lr": 3e-05, "grad_norm": 0.3556979921311017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141820691943, "tokens_per_second": 151478.29184310292, "data_wait_s": 1.683599839452654e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 969000, "loss": 0.4934856593608856, "lr": 3e-05, "grad_norm": 0.3608491779152761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141835475142, "tokens_per_second": 153380.01690895387, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51bb6395d1489f6d4bce99cd810b2bb2ec667a82e4a1d99a492b5628a9c9f6d3
-size 936231
+oid sha256:56e0fa822b6c6282842060746e65d1e307c608d10db214b791b4ffb5497ce1de
+size 938349