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": 76783.56477606195,
-      "required_pages_per_second_upper_estimate": 0.005312065173624237
+      "required_token_positions_per_second": 75924.27693677116,
+      "required_pages_per_second_upper_estimate": 0.0052526176470796175
     },
     {
       "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": 9597.945597007743,
-      "required_pages_per_second_upper_estimate": 0.004018459393571661
+      "required_token_positions_per_second": 9490.534617096395,
+      "required_pages_per_second_upper_estimate": 0.003973488659279099
     },
     {
       "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": 19195.891194015487,
-      "required_pages_per_second_upper_estimate": 0.003997997906020167
+      "required_token_positions_per_second": 18981.06923419279,
+      "required_pages_per_second_upper_estimate": 0.00395325615702515
     },
     {
       "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": 28793.836791023226,
-      "required_pages_per_second_upper_estimate": 0.001726146549978756
+      "required_token_positions_per_second": 28471.603851289186,
+      "required_pages_per_second_upper_estimate": 0.00170682917726291
     },
     {
       "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": 28793.836791023226,
-      "required_pages_per_second_upper_estimate": 0.002238594159798871
+      "required_token_positions_per_second": 28471.603851289186,
+      "required_pages_per_second_upper_estimate": 0.0022135419660874608
     },
     {
       "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": 28793.836791023226,
-      "required_pages_per_second_upper_estimate": 0.36933642194203803
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28471.603851289186,
+      "required_pages_per_second_upper_estimate": 0.3652031637779042
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.16362852624217794,
-    "producer_p95_seconds": 0.15412799900013852,
-    "gpu_update_seconds": 0.9602054842729242,
-    "producer_batches_per_second": 6.111403817937911,
-    "gpu_batches_per_second": 1.0414437496753193,
-    "producer_headroom": 5.86820346259047,
+    "producer_batch_seconds": 0.07834885901565514,
+    "producer_p95_seconds": 0.15171602400005213,
+    "gpu_update_seconds": 0.9710727974584441,
+    "producer_batches_per_second": 12.763427732881047,
+    "gpu_batches_per_second": 1.0297889124453554,
+    "producer_headroom": 12.394217473727485,
     "sustainable": true,
-    "buffer_seconds": 60.49294550919422,
+    "buffer_seconds": 60.206513442423535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450794.5539026535,
+    "generation_bytes_per_second": 449052.8702342743,
     "upload_bytes_per_second": 5650181.368266754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.73001730737404,
+    "estimated_full_shards_per_hour": 96.35629253646061,
     "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": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 560600, "loss": 0.541042685508728, "lr": 0.0003, "grad_norm": 0.1687460609301299, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82054358655, "tokens_per_second": 153677.74438393157, "data_wait_s": 1.1931000244658208e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560700, "loss": 0.6451371908187866, "lr": 0.0003, "grad_norm": 0.17166284654148692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82069152100, "tokens_per_second": 153397.27259784646, "data_wait_s": 1.426699964213185e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560800, "loss": 0.5536489486694336, "lr": 0.0003, "grad_norm": 0.16716914098514782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82083838576, "tokens_per_second": 152964.54929647542, "data_wait_s": 1.4391000149771571e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 560900, "loss": 0.6532272100448608, "lr": 0.0003, "grad_norm": 0.225882472750135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82098261619, "tokens_per_second": 149525.24728713714, "data_wait_s": 1.5470999642275274e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e729996b5a432f9fe12db13ba5b6c13d5efc76f2fdf0a64717b67ce1fac8e7ca
-size 79314
+oid sha256:dedaf9643bb465c8d34770ee15ba1433f706065b6dc890dd074378355d2c122d
+size 81432