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": 76625.74472237298,
-      "required_pages_per_second_upper_estimate": 0.005301146816117058
+      "required_token_positions_per_second": 75656.33911680247,
+      "required_pages_per_second_upper_estimate": 0.005234081086992788
     },
     {
       "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": 9578.218090296623,
-      "required_pages_per_second_upper_estimate": 0.0040101998984689
+      "required_token_positions_per_second": 9457.042389600309,
+      "required_pages_per_second_upper_estimate": 0.003959466163023729
     },
     {
       "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": 19156.436180593246,
-      "required_pages_per_second_upper_estimate": 0.003989780467222989
+      "required_token_positions_per_second": 18914.084779200617,
+      "required_pages_per_second_upper_estimate": 0.003939305061549151
     },
     {
       "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": 28734.65427088987,
-      "required_pages_per_second_upper_estimate": 0.001722598648263237
+      "required_token_positions_per_second": 28371.127168800926,
+      "required_pages_per_second_upper_estimate": 0.00170080575356675
     },
     {
       "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": 28734.65427088987,
-      "required_pages_per_second_upper_estimate": 0.0022339929791748976
+      "required_token_positions_per_second": 28371.127168800926,
+      "required_pages_per_second_upper_estimate": 0.002205730345974218
     },
     {
       "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": 28734.65427088987,
-      "required_pages_per_second_upper_estimate": 0.3685772921190066
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28371.127168800926,
+      "required_pages_per_second_upper_estimate": 0.36391435677840106
     }
   ],
   "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": 0.09721103403148845,
-    "producer_p95_seconds": 0.35396061599749373,
-    "gpu_update_seconds": 0.9621831444135132,
-    "producer_batches_per_second": 10.286898086857933,
-    "gpu_batches_per_second": 1.0393031782005884,
-    "producer_headroom": 9.897879947474319,
+    "producer_batch_seconds": 0.17910589928141007,
+    "producer_p95_seconds": 0.3499089239994646,
+    "gpu_update_seconds": 0.9745118632580756,
+    "producer_batches_per_second": 5.5832890151139365,
+    "gpu_batches_per_second": 1.0261547731771168,
+    "producer_headroom": 5.4409813812270285,
     "sustainable": true,
-    "buffer_seconds": 60.61753809805133,
+    "buffer_seconds": 60.419735522000686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333456.92247299606,
-    "upload_bytes_per_second": 5046922.71974306,
+    "generation_bytes_per_second": 342613.60773671675,
+    "upload_bytes_per_second": 5079556.566866403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.55209308283244,
+    "estimated_full_shards_per_hour": 73.5169045837033,
     "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": 1.3692391859985946,
-      "bytes": 8054475,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.189110659997823,
       "bytes": 6011923,
@@ -562,6 +556,12 @@
       "bytes": 8105432,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.5261870840040501,
+      "bytes": 11600268,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -401,3 +401,4 @@
 {"step": 965100, "loss": 0.5409970879554749, "lr": 3e-05, "grad_norm": 0.34271652718062795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141264249572, "tokens_per_second": 131399.95456990355, "data_wait_s": 1.9553997844923288e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 965200, "loss": 0.5364992022514343, "lr": 3e-05, "grad_norm": 0.3412135978278283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141279198323, "tokens_per_second": 154604.31451173962, "data_wait_s": 1.1733995052054524e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 965300, "loss": 0.5464724898338318, "lr": 3e-05, "grad_norm": 0.3479924102645416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141293728848, "tokens_per_second": 150881.69473970743, "data_wait_s": 1.3823002518620342e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 965400, "loss": 0.5208156108856201, "lr": 3e-05, "grad_norm": 0.35244522097223124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141308270148, "tokens_per_second": 151777.1182755648, "data_wait_s": 1.0207993909716606e-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 sha256:7c791a6622ccd8ef4cada5dd5626647b422db76ac2310bd70412f04fa7588342
-size 859983
+oid sha256:612620167799ec0321e96a32dea4c6629699cc3045ebe1688e3124042b2c9096
+size 862101