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": 75645.3578037919,
-      "required_pages_per_second_upper_estimate": 0.006556747893026131
+      "required_token_positions_per_second": 75569.80320427881,
+      "required_pages_per_second_upper_estimate": 0.0065501990118317175
     },
     {
       "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": 9455.669725473988,
-      "required_pages_per_second_upper_estimate": 0.0037720080283524765
+      "required_token_positions_per_second": 9446.225400534851,
+      "required_pages_per_second_upper_estimate": 0.0037682405459290138
     },
     {
       "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": 18911.339450947977,
-      "required_pages_per_second_upper_estimate": 0.004374805553769972
+      "required_token_positions_per_second": 18892.450801069703,
+      "required_pages_per_second_upper_estimate": 0.004370435997049517
     },
     {
       "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": 28367.009176421965,
-      "required_pages_per_second_upper_estimate": 0.0017274020286491526
+      "required_token_positions_per_second": 28338.676201604554,
+      "required_pages_per_second_upper_estimate": 0.0017256766991344027
     },
     {
       "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": 28367.009176421965,
-      "required_pages_per_second_upper_estimate": 0.0021837708913081807
+      "required_token_positions_per_second": 28338.676201604554,
+      "required_pages_per_second_upper_estimate": 0.0021815897404760423
     },
     {
       "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": 28367.009176421965,
-      "required_pages_per_second_upper_estimate": 0.3522233156986474
+      "required_token_positions_per_second": 28338.676201604554,
+      "required_pages_per_second_upper_estimate": 0.351871514975782
     }
   ],
   "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": 0.06372443303902742,
-    "producer_p95_seconds": 0.2250764780001191,
-    "gpu_update_seconds": 0.9746533315532048,
-    "producer_batches_per_second": 15.6925680199863,
-    "gpu_batches_per_second": 1.0260058295870214,
-    "producer_headroom": 15.294813701304925,
+    "producer_batch_seconds": 0.11870845573427857,
+    "producer_p95_seconds": 0.27004806399963854,
+    "gpu_update_seconds": 0.9756277887968018,
+    "producer_batches_per_second": 8.423999737966748,
+    "gpu_batches_per_second": 1.0249810547455351,
+    "producer_headroom": 8.218688237177336,
     "sustainable": true,
-    "buffer_seconds": 60.42850655629869,
+    "buffer_seconds": 60.48892290540171,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282493.3255318214,
+    "generation_bytes_per_second": 284286.995921983,
     "upload_bytes_per_second": 5952913.5623308495,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.61649155107481,
+    "estimated_full_shards_per_hour": 61.00137146229379,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 129900, "loss": 0.603065631352365, "lr": 0.0003, "grad_norm": 0.1387297366005279, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19010098040, "tokens_per_second": 134676.9472382548, "data_wait_s": 7.50599974708166e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 130000, "loss": 0.6065531261265278, "lr": 0.0003, "grad_norm": 0.15030424920394553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19024827314, "tokens_per_second": 150786.94044244278, "data_wait_s": 1.2477999916882254e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 130100, "loss": 0.6317621618509293, "lr": 0.0003, "grad_norm": 0.15858009652001684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19039552814, "tokens_per_second": 141480.0370326361, "data_wait_s": 1.3619999663205817e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 130200, "loss": 0.5923965331166983, "lr": 0.0003, "grad_norm": 0.13168526747637468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19054230913, "tokens_per_second": 150204.11931707352, "data_wait_s": 9.613000656827353e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24ad53b14078f559767776d418d363335d39155566f3c589404aee6bbedbb182
-size 322332
+oid sha256:0b66d668c24212dd951ca72b579dc18360eb7487235922253087545d734acd50
+size 324450