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": 75897.80758020072,
-      "required_pages_per_second_upper_estimate": 0.006578629599817139
+      "required_token_positions_per_second": 75111.59308802226,
+      "required_pages_per_second_upper_estimate": 0.006510482520277522
     },
     {
       "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": 9487.22594752509,
-      "required_pages_per_second_upper_estimate": 0.0037845962771362257
+      "required_token_positions_per_second": 9388.949136002782,
+      "required_pages_per_second_upper_estimate": 0.003745392187650703
     },
     {
       "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": 18974.45189505018,
-      "required_pages_per_second_upper_estimate": 0.004389405506971937
+      "required_token_positions_per_second": 18777.898272005565,
+      "required_pages_per_second_upper_estimate": 0.004343936285506184
     },
     {
       "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": 28461.677842575275,
-      "required_pages_per_second_upper_estimate": 0.0017331668537297839
+      "required_token_positions_per_second": 28166.847408008347,
+      "required_pages_per_second_upper_estimate": 0.0017152132271204008
     },
     {
       "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": 28461.677842575275,
-      "required_pages_per_second_upper_estimate": 0.002191058747288303
+      "required_token_positions_per_second": 28166.847408008347,
+      "required_pages_per_second_upper_estimate": 0.0021683618842924628
     },
     {
       "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": 28461.677842575275,
-      "required_pages_per_second_upper_estimate": 0.3533987836966273
+      "required_token_positions_per_second": 28166.847408008347,
+      "required_pages_per_second_upper_estimate": 0.34973797643329585
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05422459449211914,
-    "producer_p95_seconds": 0.19416267099950346,
-    "gpu_update_seconds": 0.9714114590476424,
-    "producer_batches_per_second": 18.44181610515017,
-    "gpu_batches_per_second": 1.0294298988199968,
-    "producer_headroom": 17.914591490192237,
+    "producer_batch_seconds": 0.4275812469297051,
+    "producer_p95_seconds": 0.20754633199976524,
+    "gpu_update_seconds": 0.9815795001658286,
+    "producer_batches_per_second": 2.3387368065849747,
+    "gpu_batches_per_second": 1.0187661822919687,
+    "producer_headroom": 2.2956561056271054,
     "sustainable": true,
-    "buffer_seconds": 60.22751046095382,
+    "buffer_seconds": 60.857929010281374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320580.3316713517,
-    "upload_bytes_per_second": 4929764.27135805,
+    "generation_bytes_per_second": 289225.8327437046,
+    "upload_bytes_per_second": 4876013.597572228,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.78907644849217,
+    "estimated_full_shards_per_hour": 62.0611308739982,
     "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": [
@@ -412,6 +412,12 @@
       "bytes": 7407097,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.7928119780008274,
+      "bytes": 1141631,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -76,3 +76,4 @@
 {"step": 162600, "loss": 0.5990036465227604, "lr": 0.0003, "grad_norm": 0.12867135792371365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23797449426, "tokens_per_second": 149717.1590024211, "data_wait_s": 1.1140999959025066e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162700, "loss": 0.5720186643302441, "lr": 0.0003, "grad_norm": 0.1345017926209548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23812088597, "tokens_per_second": 149943.13882640048, "data_wait_s": 1.2315998901613057e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 162800, "loss": 0.614055659621954, "lr": 0.0003, "grad_norm": 0.1426784637081092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23826817474, "tokens_per_second": 148782.67257115056, "data_wait_s": 9.510999007034115e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 162900, "loss": 0.6072080656886101, "lr": 0.0003, "grad_norm": 0.153760050164428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23841652859, "tokens_per_second": 151678.14044052552, "data_wait_s": 1.113600046664942e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c7a892e96bfbd2c74f5dacf93f73c1eec2612614366408bf4c34e2f00362a55
-size 166152
+oid sha256:8a0d3fe49ffb5f50d2972133ecf77585a7d5b4e820f5bcbb2f678768415b6451
+size 168270