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": 76163.98598798746,
-      "required_pages_per_second_upper_estimate": 0.0066017012695810855
+      "required_token_positions_per_second": 75314.09942689793,
+      "required_pages_per_second_upper_estimate": 0.006528035256483641
     },
     {
       "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": 9520.498248498432,
-      "required_pages_per_second_upper_estimate": 0.0037978690954597223
+      "required_token_positions_per_second": 9414.26242836224,
+      "required_pages_per_second_upper_estimate": 0.003755490038440339
     },
     {
       "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": 19040.996496996864,
-      "required_pages_per_second_upper_estimate": 0.004404799429487309
+      "required_token_positions_per_second": 18828.52485672448,
+      "required_pages_per_second_upper_estimate": 0.004355647854883446
     },
     {
       "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": 28561.494745495293,
-      "required_pages_per_second_upper_estimate": 0.0017392451794187968
+      "required_token_positions_per_second": 28242.787285086724,
+      "required_pages_per_second_upper_estimate": 0.0017198375672087308
     },
     {
       "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": 28561.494745495293,
-      "required_pages_per_second_upper_estimate": 0.002198742928785958
+      "required_token_positions_per_second": 28242.787285086724,
+      "required_pages_per_second_upper_estimate": 0.0021742079462449904
     },
     {
       "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": 28561.494745495293,
-      "required_pages_per_second_upper_estimate": 0.3546381755652097
+      "required_token_positions_per_second": 28242.787285086724,
+      "required_pages_per_second_upper_estimate": 0.35068089555218995
     }
   ],
   "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": 0.09076273605472807,
-    "producer_p95_seconds": 0.26034281300235307,
-    "gpu_update_seconds": 0.9680165637815797,
-    "producer_batches_per_second": 11.017737493028202,
-    "gpu_batches_per_second": 1.03304017453325,
-    "producer_headroom": 10.665352388648637,
+    "producer_batch_seconds": 0.34995511625766085,
+    "producer_p95_seconds": 0.3382720430017798,
+    "gpu_update_seconds": 0.978940205898134,
+    "producer_batches_per_second": 2.8575093020321263,
+    "gpu_batches_per_second": 1.0215128502997224,
+    "producer_headroom": 2.7973307444871627,
     "sustainable": true,
-    "buffer_seconds": 60.01702695445795,
+    "buffer_seconds": 60.69429276568431,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436315.51467400003,
+    "generation_bytes_per_second": 418339.1527070613,
     "upload_bytes_per_second": 3412984.621486847,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.62315254368782,
+    "estimated_full_shards_per_hour": 89.76584373387222,
     "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

@@ -263,3 +263,4 @@
 {"step": 66300, "loss": 0.6145434500649571, "lr": 0.0003, "grad_norm": 0.14590196705388214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9703281602, "tokens_per_second": 147787.86374603992, "data_wait_s": 1.2303997209528461e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66400, "loss": 0.6198974773287773, "lr": 0.0003, "grad_norm": 0.13519091249670845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9717852472, "tokens_per_second": 149328.9064931025, "data_wait_s": 9.00899976841174e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 66500, "loss": 0.5951087810099125, "lr": 0.0003, "grad_norm": 0.13952694373766053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9732476774, "tokens_per_second": 148525.42918499935, "data_wait_s": 1.1972999345744029e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 66600, "loss": 0.5865750210359693, "lr": 0.0003, "grad_norm": 0.1314899339702683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9747201222, "tokens_per_second": 150907.16278123777, "data_wait_s": 1.5180001355474815e-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 sha256:29c604ba9882695b4120453d109891b7092e0c69bd25fd430bdb2701cffb365f
-size 565350
+oid sha256:821f133240065728164ff288715fd3a652891956ff1879a0c725b679a49ba8f7
+size 567468