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": 77436.89032736821,
-      "required_pages_per_second_upper_estimate": 0.005357263751187736
+      "required_token_positions_per_second": 77572.96376180205,
+      "required_pages_per_second_upper_estimate": 0.005366677627115737
     },
     {
       "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": 9679.611290921026,
-      "required_pages_per_second_upper_estimate": 0.004052651114239539
+      "required_token_positions_per_second": 9696.620470225256,
+      "required_pages_per_second_upper_estimate": 0.004059772502422166
     },
     {
       "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": 19359.222581842052,
-      "required_pages_per_second_upper_estimate": 0.004032015526766088
+      "required_token_positions_per_second": 19393.240940450512,
+      "required_pages_per_second_upper_estimate": 0.004039100653739785
     },
     {
       "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": 29038.833872763076,
-      "required_pages_per_second_upper_estimate": 0.0017408337509401764
+      "required_token_positions_per_second": 29089.86141067577,
+      "required_pages_per_second_upper_estimate": 0.0017438927739235035
     },
     {
       "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": 29038.833872763076,
-      "required_pages_per_second_upper_estimate": 0.0022576416052758693
+      "required_token_positions_per_second": 29089.86141067577,
+      "required_pages_per_second_upper_estimate": 0.002261608771902162
     },
     {
       "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": 29038.833872763076,
-      "required_pages_per_second_upper_estimate": 0.3724789814492256
+      "required_token_positions_per_second": 29089.86141067577,
+      "required_pages_per_second_upper_estimate": 0.3731335079164681
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06805545011718905,
-    "producer_p95_seconds": 0.15329203400006008,
-    "gpu_update_seconds": 0.9521043483062311,
-    "producer_batches_per_second": 14.693900316257341,
-    "gpu_batches_per_second": 1.0503050445877848,
-    "producer_headroom": 13.99012638468692,
+    "producer_batch_seconds": 0.04559602843748678,
+    "producer_p95_seconds": 0.1422521310000775,
+    "gpu_update_seconds": 0.9504342289459441,
+    "producer_batches_per_second": 21.9317347205146,
+    "gpu_batches_per_second": 1.0521506586615947,
+    "producer_headroom": 20.84467137853929,
     "sustainable": true,
-    "buffer_seconds": 60.93467829159879,
+    "buffer_seconds": 60.82779065254042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437161.9732586011,
-    "upload_bytes_per_second": 4824156.882697495,
+    "generation_bytes_per_second": 435721.3006613159,
+    "upload_bytes_per_second": 4880671.794164952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.80478285139584,
+    "estimated_full_shards_per_hour": 93.49564804916008,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 12779121,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.091388643999835,
+      "bytes": 6015354,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 887300, "loss": 0.541535496711731, "lr": 3e-05, "grad_norm": 0.30549372524149293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129876072345, "tokens_per_second": 148892.38539015697, "data_wait_s": 1.0383000244473806e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887400, "loss": 0.5499482750892639, "lr": 3e-05, "grad_norm": 0.33402700638135396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129890689891, "tokens_per_second": 152637.59663709343, "data_wait_s": 1.0358000054111471e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 887500, "loss": 0.4985582232475281, "lr": 3e-05, "grad_norm": 0.30761058333817004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129905366636, "tokens_per_second": 153305.5516986935, "data_wait_s": 1.6431999938504305e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 887600, "loss": 0.49373859167099, "lr": 3e-05, "grad_norm": 0.29812396705866234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129920081715, "tokens_per_second": 153026.20680727615, "data_wait_s": 8.690999948157696e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5efbde38842705396c8ad76ccb4dd0cd155e1aa7fdce87c709f9298398b7cbc2
-size 53115
+oid sha256:97afbe166eff3e406c4b857935f0ebe457d628cebe533998554b699bd627395d
+size 55233