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": 75391.38912180909,
-      "required_pages_per_second_upper_estimate": 0.0065347345313496
+      "required_token_positions_per_second": 75794.0162882829,
+      "required_pages_per_second_upper_estimate": 0.006569633233690326
     },
     {
       "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": 9423.923640226136,
-      "required_pages_per_second_upper_estimate": 0.003759344040300836
+      "required_token_positions_per_second": 9474.252036035363,
+      "required_pages_per_second_upper_estimate": 0.0037794207898657105
     },
     {
       "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": 18847.84728045227,
-      "required_pages_per_second_upper_estimate": 0.004360117757549827
+      "required_token_positions_per_second": 18948.504072070726,
+      "required_pages_per_second_upper_estimate": 0.004383402934791728
     },
     {
       "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": 28271.770920678406,
-      "required_pages_per_second_upper_estimate": 0.0017216025185508806
+      "required_token_positions_per_second": 28422.75610810609,
+      "required_pages_per_second_upper_estimate": 0.0017307967243071693
     },
     {
       "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": 28271.770920678406,
-      "required_pages_per_second_upper_estimate": 0.002176439186744676
+      "required_token_positions_per_second": 28422.75610810609,
+      "required_pages_per_second_upper_estimate": 0.0021880624444266085
     },
     {
       "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": 28271.770920678406,
-      "required_pages_per_second_upper_estimate": 0.35104077530425026
+      "required_token_positions_per_second": 28422.75610810609,
+      "required_pages_per_second_upper_estimate": 0.35291550601718574
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.06322811708594855,
-    "producer_p95_seconds": 0.14520624199985832,
-    "gpu_update_seconds": 0.9779366166191531,
-    "producer_batches_per_second": 15.815748532265468,
-    "gpu_batches_per_second": 1.022561158878704,
-    "producer_headroom": 15.466799608943028,
+    "producer_batch_seconds": 0.044901345335947696,
+    "producer_p95_seconds": 0.14561744000002363,
+    "gpu_update_seconds": 0.9727416966475981,
+    "producer_batches_per_second": 22.27104761601446,
+    "gpu_batches_per_second": 1.028022139326754,
+    "producer_headroom": 21.66397664412135,
     "sustainable": true,
-    "buffer_seconds": 60.63207023038749,
+    "buffer_seconds": 60.30998519215108,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 63,
+    "upload_pages_per_commit": 67,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284343.23951015924,
-    "upload_bytes_per_second": 5026525.805553968,
+    "generation_bytes_per_second": 284739.20211454737,
+    "upload_bytes_per_second": 4791606.172962086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.013440027032686,
+    "estimated_full_shards_per_hour": 61.0984043843967,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 10663078,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.8003151779998916,
+      "bytes": 1078242,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 316100, "loss": 0.5824047094210982, "lr": 0.0003, "grad_norm": 0.15162778451768252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46268355495, "tokens_per_second": 129192.18210135636, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 316200, "loss": 0.5630151983350515, "lr": 0.0003, "grad_norm": 0.18497926545476986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46283079553, "tokens_per_second": 149341.77200001292, "data_wait_s": 1.1314999937894754e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 316300, "loss": 0.5828749593347311, "lr": 0.0003, "grad_norm": 0.16754748799639235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46297730448, "tokens_per_second": 149039.69755937686, "data_wait_s": 9.000000090964022e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 316400, "loss": 0.581691725179553, "lr": 0.0003, "grad_norm": 0.16576541745061205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46312342500, "tokens_per_second": 148523.02108168503, "data_wait_s": 1.0182000096392585e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6dc38238b1d7e14188d506cfacb803436b1f037993e9d3de57acefd33dc0db36
-size 27699
+oid sha256:85a2cff5b6678f3972b3cec4515117203a9bfcd5c8a0bf5c6083830ffc52cc95
+size 29817