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": 76971.52490069767,
-      "required_pages_per_second_upper_estimate": 0.0053250686911741115
+      "required_token_positions_per_second": 75920.89330740352,
+      "required_pages_per_second_upper_estimate": 0.005252383559748862
     },
     {
       "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": 9621.440612587208,
-      "required_pages_per_second_upper_estimate": 0.004028296265963066
+      "required_token_positions_per_second": 9490.11166342544,
+      "required_pages_per_second_upper_estimate": 0.003973311577409347
     },
     {
       "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": 19242.881225174417,
-      "required_pages_per_second_upper_estimate": 0.004007784690300118
+      "required_token_positions_per_second": 18980.22332685088,
+      "required_pages_per_second_upper_estimate": 0.003953079976833907
     },
     {
       "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": 28864.321837761625,
-      "required_pages_per_second_upper_estimate": 0.0017303720209062882
+      "required_token_positions_per_second": 28470.334990276322,
+      "required_pages_per_second_upper_estimate": 0.0017067531109826275
     },
     {
       "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": 28864.321837761625,
-      "required_pages_per_second_upper_estimate": 0.002244074062151826
+      "required_token_positions_per_second": 28470.334990276322,
+      "required_pages_per_second_upper_estimate": 0.0022134433177248406
     },
     {
       "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": 28864.321837761625,
-      "required_pages_per_second_upper_estimate": 0.3702405284406514
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28470.334990276322,
+      "required_pages_per_second_upper_estimate": 0.36518688819122797
     }
   ],
   "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": 0.48976437372647297,
-    "producer_p95_seconds": 0.23143664199960767,
-    "gpu_update_seconds": 0.9578607166106922,
-    "producer_batches_per_second": 2.0417981659043396,
-    "gpu_batches_per_second": 1.0439931220255216,
-    "producer_headroom": 1.9557582543675276,
+    "producer_batch_seconds": 0.14551501082019058,
+    "producer_p95_seconds": 0.24350384499848587,
+    "gpu_update_seconds": 0.971116076064536,
+    "producer_batches_per_second": 6.872143254249392,
+    "gpu_batches_per_second": 1.0297430190348784,
+    "producer_headroom": 6.673648791220041,
     "sustainable": true,
-    "buffer_seconds": 60.34522514647361,
+    "buffer_seconds": 60.20919671600123,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263502.6407069717,
-    "upload_bytes_per_second": 4802558.572929637,
+    "generation_bytes_per_second": 281853.3535732672,
+    "upload_bytes_per_second": 4822793.520367699,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.54153266817916,
+    "estimated_full_shards_per_hour": 60.479168466553794,
     "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": 1.7708508949999668,
-      "bytes": 6504956,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1802944610008126,
       "bytes": 6153792,
@@ -562,6 +556,12 @@
       "bytes": 11841247,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3097548330006248,
+      "bytes": 6099339,
+      "pages": 40,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -207,3 +207,4 @@
 {"step": 945700, "loss": 0.5435240268707275, "lr": 3e-05, "grad_norm": 0.3412607514844836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138424015138, "tokens_per_second": 152529.5826100117, "data_wait_s": 1.0265001037623733e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945800, "loss": 0.5362918972969055, "lr": 3e-05, "grad_norm": 0.35337433595240636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138438740144, "tokens_per_second": 148556.058747488, "data_wait_s": 1.22519995784387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945900, "loss": 0.4961773753166199, "lr": 3e-05, "grad_norm": 0.3491370009812067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138453339381, "tokens_per_second": 152265.35082931517, "data_wait_s": 8.971001079771668e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 946000, "loss": 0.5332742929458618, "lr": 3e-05, "grad_norm": 0.340665019120442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138467927763, "tokens_per_second": 152139.04841670825, "data_wait_s": 1.3922999642090872e-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 sha256:7376dda278bccc34237b94aef06b08cc008da3ce03416a2ee70508fb050ab0f9
-size 445959
+oid sha256:37a7404fea90db5b7045a3b19aa668ac1b50850be016747cfa0724c46307582a
+size 448077