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": 75476.69250524513,
-      "required_pages_per_second_upper_estimate": 0.006542128412426387
+      "required_token_positions_per_second": 75724.69916464922,
+      "required_pages_per_second_upper_estimate": 0.0065636250010964025
     },
     {
       "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": 9434.58656315564,
-      "required_pages_per_second_upper_estimate": 0.0037635976396823206
+      "required_token_positions_per_second": 9465.587395581153,
+      "required_pages_per_second_upper_estimate": 0.0037759643352406067
     },
     {
       "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": 18869.17312631128,
-      "required_pages_per_second_upper_estimate": 0.004365051116667241
+      "required_token_positions_per_second": 18931.174791162306,
+      "required_pages_per_second_upper_estimate": 0.004379394110638492
     },
     {
       "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": 28303.759689466922,
-      "required_pages_per_second_upper_estimate": 0.001723550466738002
+      "required_token_positions_per_second": 28396.762186743457,
+      "required_pages_per_second_upper_estimate": 0.0017292138308757504
     },
     {
       "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": 28303.759689466922,
-      "required_pages_per_second_upper_estimate": 0.002178901770716596
+      "required_token_positions_per_second": 28396.762186743457,
+      "required_pages_per_second_upper_estimate": 0.002186061360404333
     },
     {
       "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": 28303.759689466922,
-      "required_pages_per_second_upper_estimate": 0.3514379687530815
+      "required_token_positions_per_second": 28396.762186743457,
+      "required_pages_per_second_upper_estimate": 0.3525927485099204
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.43243930066404346,
-    "producer_p95_seconds": 0.35765524900125456,
-    "gpu_update_seconds": 0.9768313575065096,
-    "producer_batches_per_second": 2.3124632716416476,
-    "gpu_batches_per_second": 1.0237181600646312,
-    "producer_headroom": 2.258886636821655,
+    "producer_batch_seconds": 0.10487522683607153,
+    "producer_p95_seconds": 0.34487755699956324,
+    "gpu_update_seconds": 0.9736321281342066,
+    "producer_batches_per_second": 9.535140282109529,
+    "gpu_batches_per_second": 1.0270819656663577,
+    "producer_headroom": 9.2837189249285,
     "sustainable": true,
-    "buffer_seconds": 60.5635441654036,
+    "buffer_seconds": 60.36519194432081,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318343.5095025213,
+    "generation_bytes_per_second": 323874.9203211429,
     "upload_bytes_per_second": 3391082.3296538573,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3091064816163,
+    "estimated_full_shards_per_hour": 69.49601847863879,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -284,3 +284,4 @@
 {"step": 68400, "loss": 0.6469593495130539, "lr": 0.0003, "grad_norm": 0.12514281301613506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10010732696, "tokens_per_second": 150178.65490609565, "data_wait_s": 1.562300167279318e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 68500, "loss": 0.633449487388134, "lr": 0.0003, "grad_norm": 0.13704763604633602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10025437145, "tokens_per_second": 149418.99525996717, "data_wait_s": 9.718001820147038e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 68600, "loss": 0.604379765689373, "lr": 0.0003, "grad_norm": 0.13055077179418748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10039940722, "tokens_per_second": 147720.23354195536, "data_wait_s": 1.296499976888299e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 68700, "loss": 0.5999638177454472, "lr": 0.0003, "grad_norm": 0.13280039697618273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10054465618, "tokens_per_second": 148837.63019668384, "data_wait_s": 1.3896999007556587e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e11d5686424b226b70d4488c0c650910047d849d80d69e8d224de4228fab0893
-size 609828
+oid sha256:f2ee9d2cf18a4696bffab9362d205e4fdd58cbaa8ce6c65a50d9abf1492afeb3
+size 611946