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": 77077.71212277717,
-      "required_pages_per_second_upper_estimate": 0.005332414969585872
+      "required_token_positions_per_second": 77223.05774441108,
+      "required_pages_per_second_upper_estimate": 0.005342470316938812
     },
     {
       "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": 9634.714015347146,
-      "required_pages_per_second_upper_estimate": 0.004033853562518483
+      "required_token_positions_per_second": 9652.882218051385,
+      "required_pages_per_second_upper_estimate": 0.004041460209595533
     },
     {
       "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": 19269.428030694293,
-      "required_pages_per_second_upper_estimate": 0.004013313689803568
+      "required_token_positions_per_second": 19305.76443610277,
+      "required_pages_per_second_upper_estimate": 0.004020881604794703
     },
     {
       "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": 28904.14204604144,
-      "required_pages_per_second_upper_estimate": 0.0017327591815907257
+      "required_token_positions_per_second": 28958.646654154152,
+      "required_pages_per_second_upper_estimate": 0.0017360266496233648
     },
     {
       "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": 28904.14204604144,
-      "required_pages_per_second_upper_estimate": 0.002247169908194996
+      "required_token_positions_per_second": 28958.646654154152,
+      "required_pages_per_second_upper_estimate": 0.0022514074017353247
     },
     {
       "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": 28904.14204604144,
-      "required_pages_per_second_upper_estimate": 0.3707512993168564
+      "required_token_positions_per_second": 28958.646654154152,
+      "required_pages_per_second_upper_estimate": 0.37145042590723765
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.34663867851583063,
-    "producer_p95_seconds": 0.26674412499778555,
-    "gpu_update_seconds": 0.9565411059757273,
-    "producer_batches_per_second": 2.8848482929879706,
-    "gpu_batches_per_second": 1.0454333784013832,
-    "producer_headroom": 2.7594759767469026,
+    "producer_batch_seconds": 0.07235783015624975,
+    "producer_p95_seconds": 0.26276853599847527,
+    "gpu_update_seconds": 0.9547407491169433,
+    "producer_batches_per_second": 13.820204362687447,
+    "gpu_batches_per_second": 1.0474047545628673,
+    "producer_headroom": 13.19471226618146,
     "sustainable": true,
-    "buffer_seconds": 60.26208967647082,
+    "buffer_seconds": 60.14866719436743,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411361.3399946071,
-    "upload_bytes_per_second": 4915333.535647577,
+    "generation_bytes_per_second": 411129.12935704755,
+    "upload_bytes_per_second": 4930776.978256061,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.26856756094608,
+    "estimated_full_shards_per_hour": 88.21874056371279,
     "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": 1.2794374899985996,
-      "bytes": 6045695,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.710491572997853,
       "bytes": 17528313,
@@ -562,6 +556,12 @@
       "bytes": 5736433,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.218638172998908,
+      "bytes": 7112834,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 909200, "loss": 0.6002414226531982, "lr": 3e-05, "grad_norm": 0.3585365651129577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133080864528, "tokens_per_second": 153427.0324208804, "data_wait_s": 1.3517001207219437e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909300, "loss": 0.5651774406433105, "lr": 3e-05, "grad_norm": 0.32240538373881356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133095384685, "tokens_per_second": 151273.27839818533, "data_wait_s": 1.6969999705906957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909400, "loss": 0.500087559223175, "lr": 3e-05, "grad_norm": 0.31723700030761487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133110124186, "tokens_per_second": 153149.113347913, "data_wait_s": 1.7561000277055427e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 909500, "loss": 0.5202333331108093, "lr": 3e-05, "grad_norm": 0.3383423403843563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133124866204, "tokens_per_second": 153046.82829895406, "data_wait_s": 1.6049001715146005e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4ba7769180a30f72c4bd7b6211121aea900b1ac1491a40a39f5cc8861396eb4
-size 520089
+oid sha256:05e0f28fa338f0f66e91e4c0e1b6099a8412cdf838da8d18671101667954c2ee
+size 522207