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": 75745.03985089272,
-      "required_pages_per_second_upper_estimate": 0.006565388080226991
+      "required_token_positions_per_second": 75767.41078337622,
+      "required_pages_per_second_upper_estimate": 0.0065673271359560515
     },
     {
       "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": 9468.12998136159,
-      "required_pages_per_second_upper_estimate": 0.003776978610723468
+      "required_token_positions_per_second": 9470.926347922028,
+      "required_pages_per_second_upper_estimate": 0.0037780941231538327
     },
     {
       "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": 18936.25996272318,
-      "required_pages_per_second_upper_estimate": 0.004380570475583138
+      "required_token_positions_per_second": 18941.852695844056,
+      "required_pages_per_second_upper_estimate": 0.0043818642559619115
     },
     {
       "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": 28404.389944084767,
-      "required_pages_per_second_upper_estimate": 0.0017296783212780863
+      "required_token_positions_per_second": 28412.779043766084,
+      "required_pages_per_second_upper_estimate": 0.001730189173434474
     },
     {
       "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": 28404.389944084767,
-      "required_pages_per_second_upper_estimate": 0.002186648566279451
+      "required_token_positions_per_second": 28412.779043766084,
+      "required_pages_per_second_upper_estimate": 0.0021872943823954332
     },
     {
       "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": 28404.389944084767,
-      "required_pages_per_second_upper_estimate": 0.3526874597276378
+      "required_token_positions_per_second": 28412.779043766084,
+      "required_pages_per_second_upper_estimate": 0.35279162426916927
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.1472307845081673,
-    "producer_p95_seconds": 0.39345709899498615,
-    "gpu_update_seconds": 0.9733706675068976,
-    "producer_batches_per_second": 6.792057811418694,
-    "gpu_batches_per_second": 1.0273578538803807,
-    "producer_headroom": 6.611189845646051,
+    "producer_batch_seconds": 0.12493559953929889,
+    "producer_p95_seconds": 0.3801645399944391,
+    "gpu_update_seconds": 0.9730832720520565,
+    "producer_batches_per_second": 8.00412375405816,
+    "gpu_batches_per_second": 1.0276612790714006,
+    "producer_headroom": 7.788678932508504,
     "sustainable": true,
-    "buffer_seconds": 60.34898138542765,
+    "buffer_seconds": 60.3311628672275,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371388.7748888109,
-    "upload_bytes_per_second": 5211400.917541206,
+    "generation_bytes_per_second": 371381.9426136919,
+    "upload_bytes_per_second": 5175347.5918259155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.6913856029343,
+    "estimated_full_shards_per_hour": 79.689919555741,
     "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": 1.2303334819989686,
-      "bytes": 10801054,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9937779290012259,
       "bytes": 2662341,
@@ -562,6 +556,12 @@
       "bytes": 1181839,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.1708070859967847,
+      "bytes": 7634853,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 391600, "loss": 0.5928702764213085, "lr": 0.0003, "grad_norm": 0.1683905331201567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57319512166, "tokens_per_second": 148596.08297080142, "data_wait_s": 9.602998034097254e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 391700, "loss": 0.5892277006059885, "lr": 0.0003, "grad_norm": 0.16942131957179868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57334027566, "tokens_per_second": 148434.1874087957, "data_wait_s": 1.3990997103974223e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 391800, "loss": 0.5905763898044825, "lr": 0.0003, "grad_norm": 0.1508239424400319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57348573446, "tokens_per_second": 148162.10756469768, "data_wait_s": 1.5211997379083186e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 391900, "loss": 0.5476494245231152, "lr": 0.0003, "grad_norm": 0.14311585357797804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57363209465, "tokens_per_second": 131533.90772729163, "data_wait_s": 9.354000212624669e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29f75a6f13439707daf1ecc8bd64ff40c1b2ec757bb4cb79b283ed96d06d82a1
-size 785070
+oid sha256:2b5589f521d7682d5d6c5f6f43374b06e6fe2b36985ddd97f28d5ed524e0805e
+size 787188