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": 74732.41838315224,
-      "required_pages_per_second_upper_estimate": 0.006477616617868893
+      "required_token_positions_per_second": 74570.68215992746,
+      "required_pages_per_second_upper_estimate": 0.006463597731956478
     },
     {
       "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": 9341.55229789403,
-      "required_pages_per_second_upper_estimate": 0.0037264848802832417
+      "required_token_positions_per_second": 9321.335269990932,
+      "required_pages_per_second_upper_estimate": 0.003718420005581192
     },
     {
       "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": 18683.10459578806,
-      "required_pages_per_second_upper_estimate": 0.004322007436824986
+      "required_token_positions_per_second": 18642.670539981864,
+      "required_pages_per_second_upper_estimate": 0.0043126537296293
     },
     {
       "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": 28024.656893682088,
-      "required_pages_per_second_upper_estimate": 0.0017065545708139051
+      "required_token_positions_per_second": 27964.005809972794,
+      "required_pages_per_second_upper_estimate": 0.0017028612380276527
     },
     {
       "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": 28024.656893682088,
-      "required_pages_per_second_upper_estimate": 0.0021574156648910916
+      "required_token_positions_per_second": 27964.005809972794,
+      "required_pages_per_second_upper_estimate": 0.0021527465765742047
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28024.656893682088,
-      "required_pages_per_second_upper_estimate": 0.3479724461264026
+      "required_token_positions_per_second": 27964.005809972794,
+      "required_pages_per_second_upper_estimate": 0.34721936265285264
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23974966496081151,
-    "producer_p95_seconds": 0.2606211339989386,
-    "gpu_update_seconds": 0.9865598035647315,
-    "producer_batches_per_second": 4.171017299079254,
-    "gpu_batches_per_second": 1.0136232962124598,
-    "producer_headroom": 4.114958007244725,
+    "producer_batch_seconds": 0.0807801126094887,
+    "producer_p95_seconds": 0.23236994000035338,
+    "gpu_update_seconds": 0.9886995514119048,
+    "producer_batches_per_second": 12.379284550323055,
+    "gpu_batches_per_second": 1.0114296082889465,
+    "producer_headroom": 12.239393081704728,
     "sustainable": true,
-    "buffer_seconds": 60.18014801744862,
+    "buffer_seconds": 60.31067263612619,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457685.121456541,
-    "upload_bytes_per_second": 5920682.190912063,
+    "generation_bytes_per_second": 439948.7849665686,
+    "upload_bytes_per_second": 5923703.9080065815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.20857270023511,
+    "estimated_full_shards_per_hour": 94.40276777026932,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 12252293,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.9495614830011618,
+      "bytes": 5780919,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 95300, "loss": 0.625568326562643, "lr": 0.0003, "grad_norm": 0.13915032065916308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13946370067, "tokens_per_second": 148887.99585646426, "data_wait_s": 1.2813001376343891e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 95400, "loss": 0.5794907696545124, "lr": 0.0003, "grad_norm": 0.13514553789882475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13961003735, "tokens_per_second": 148911.6350202893, "data_wait_s": 8.459999662591144e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 95500, "loss": 0.6024086624383926, "lr": 0.0003, "grad_norm": 0.13020203483025394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13975402149, "tokens_per_second": 147120.69842619725, "data_wait_s": 8.13799852039665e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 95600, "loss": 0.6149090323597193, "lr": 0.0003, "grad_norm": 0.15212608340685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13989889712, "tokens_per_second": 148122.4862507203, "data_wait_s": 1.5952999092405662e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c96ae2ac269f0bddca95f7a87d885fe3b5ff105a6d431a08d86c8775ba4862db
-size 224121
+oid sha256:4f8f303fe3bce97bb9d9cf0aa29bd9c5646719c00a1db08fb164cbd2d37d0ae6
+size 226239