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": 76638.73933031264,
-      "required_pages_per_second_upper_estimate": 0.005302045813245968
+      "required_token_positions_per_second": 76615.18215859488,
+      "required_pages_per_second_upper_estimate": 0.005300416073446364
     },
     {
       "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": 9579.84241628908,
-      "required_pages_per_second_upper_estimate": 0.004010879969842158
+      "required_token_positions_per_second": 9576.89776982436,
+      "required_pages_per_second_upper_estimate": 0.00400964710785859
     },
     {
       "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": 19159.68483257816,
-      "required_pages_per_second_upper_estimate": 0.003990457075758733
+      "required_token_positions_per_second": 19153.79553964872,
+      "required_pages_per_second_upper_estimate": 0.003989230491352629
     },
     {
       "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": 28739.527248867245,
-      "required_pages_per_second_upper_estimate": 0.001722890775852374
+      "required_token_positions_per_second": 28730.693309473078,
+      "required_pages_per_second_upper_estimate": 0.0017223611946743382
     },
     {
       "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": 28739.527248867245,
-      "required_pages_per_second_upper_estimate": 0.0022343718317786735
+      "required_token_positions_per_second": 28730.693309473078,
+      "required_pages_per_second_upper_estimate": 0.00223368503184717
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28739.527248867245,
-      "required_pages_per_second_upper_estimate": 0.3686397974483042
+      "required_token_positions_per_second": 28730.693309473078,
+      "required_pages_per_second_upper_estimate": 0.36852648515890096
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06867510934377208,
-    "producer_p95_seconds": 0.16609271000015724,
-    "gpu_update_seconds": 0.9620199998623754,
-    "producer_batches_per_second": 14.561316458838471,
-    "gpu_batches_per_second": 1.0394794288508118,
-    "producer_headroom": 14.00827765772779,
+    "producer_batch_seconds": 0.04947577816407289,
+    "producer_p95_seconds": 0.1689774340002259,
+    "gpu_update_seconds": 0.962315795939526,
+    "producer_batches_per_second": 20.211910496562044,
+    "gpu_batches_per_second": 1.0391599142604557,
+    "producer_headroom": 19.45024073695756,
     "sustainable": true,
-    "buffer_seconds": 60.60725999132965,
+    "buffer_seconds": 60.62589514419014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415541.04827218805,
+    "generation_bytes_per_second": 392598.2786358123,
     "upload_bytes_per_second": 4740522.195146137,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.16543565868598,
+    "estimated_full_shards_per_hour": 84.24245137506271,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -53,3 +53,4 @@
 {"step": 402400, "loss": 0.5591439008712769, "lr": 0.0003, "grad_norm": 0.1861870593679972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58900651947, "tokens_per_second": 151104.56255020088, "data_wait_s": 1.614199936739169e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 402500, "loss": 0.6190589070320129, "lr": 0.0003, "grad_norm": 0.1705422477550829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58915397351, "tokens_per_second": 153381.7414974531, "data_wait_s": 7.648000064364169e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 402600, "loss": 0.5891031622886658, "lr": 0.0003, "grad_norm": 0.15020110006666054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58930170858, "tokens_per_second": 153633.3075546813, "data_wait_s": 1.2636999599635601e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 402700, "loss": 0.5591861009597778, "lr": 0.0003, "grad_norm": 0.17413688342588102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58944884536, "tokens_per_second": 153173.0480065034, "data_wait_s": 1.420499938831199e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9cdda879d2c2a45fa29fd7e4db13c73914043e50780a2190132da4650616d4f
-size 117438
+oid sha256:bd59fff5ba7c6a732c40b917d283a0ed37b3a99e0e20343751cff71e0568b2d2
+size 119556