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": 77149.02252247771,
-      "required_pages_per_second_upper_estimate": 0.00533734838850008
+      "required_token_positions_per_second": 77349.77256644783,
+      "required_pages_per_second_upper_estimate": 0.00535123674234621
     },
     {
       "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": 9643.627815309714,
-      "required_pages_per_second_upper_estimate": 0.0040375855844215
+      "required_token_positions_per_second": 9668.72157080598,
+      "required_pages_per_second_upper_estimate": 0.004048091815830584
     },
     {
       "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": 19287.255630619427,
-      "required_pages_per_second_upper_estimate": 0.004017026708722547
+      "required_token_positions_per_second": 19337.44314161196,
+      "required_pages_per_second_upper_estimate": 0.004027479443728623
     },
     {
       "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": 28930.88344592914,
-      "required_pages_per_second_upper_estimate": 0.0017343622876822397
+      "required_token_positions_per_second": 29006.16471241794,
+      "required_pages_per_second_upper_estimate": 0.001738875284660405
     },
     {
       "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": 28930.88344592914,
-      "required_pages_per_second_upper_estimate": 0.0022492489344132774
+      "required_token_positions_per_second": 29006.16471241794,
+      "required_pages_per_second_upper_estimate": 0.0022551017217554846
     },
     {
       "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": 28930.88344592914,
-      "required_pages_per_second_upper_estimate": 0.37109430928193765
+      "required_token_positions_per_second": 29006.16471241794,
+      "required_pages_per_second_upper_estimate": 0.3720599365377296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.24535711135138172,
-    "producer_p95_seconds": 0.32691912899463205,
-    "gpu_update_seconds": 0.9556569557121611,
-    "producer_batches_per_second": 4.075691935286426,
-    "gpu_batches_per_second": 1.0464005875987101,
-    "producer_headroom": 3.8949633472964322,
+    "producer_batch_seconds": 0.10940042345293932,
+    "producer_p95_seconds": 0.3257678099980694,
+    "gpu_update_seconds": 0.9531766875806063,
+    "producer_batches_per_second": 9.140732443601273,
+    "gpu_batches_per_second": 1.049123434332246,
+    "producer_headroom": 8.712733072652442,
     "sustainable": true,
-    "buffer_seconds": 60.206388209866155,
+    "buffer_seconds": 60.05013131757819,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409598.8519661937,
+    "generation_bytes_per_second": 425795.8908965641,
     "upload_bytes_per_second": 5026975.921434505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.89037865867003,
+    "estimated_full_shards_per_hour": 91.36588616535848,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 639700, "loss": 0.5964689254760742, "lr": 0.0001, "grad_norm": 0.1727498107872469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93635917622, "tokens_per_second": 152252.92209369232, "data_wait_s": 1.6600999515503645e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639800, "loss": 0.5595235824584961, "lr": 0.0001, "grad_norm": 0.1741859342505281, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93650638788, "tokens_per_second": 152903.79357081215, "data_wait_s": 1.479500497225672e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639900, "loss": 0.5547839403152466, "lr": 0.0001, "grad_norm": 0.1854655156901785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93665222126, "tokens_per_second": 150913.60097193083, "data_wait_s": 1.2490003427956253e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 640000, "loss": 0.5932347774505615, "lr": 0.0001, "grad_norm": 0.19791672589781517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93679740292, "tokens_per_second": 150821.95741836695, "data_wait_s": 1.6641002730466425e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07b61be8dc566ae2a9f4a4720c1a65e1fb24f5bc2f85e752b3cc48cd1cc56b23
-size 757536
+oid sha256:1682fae531048ae05cb646c1bc1718fbb9e9cef3d397de2732470502a00275a5
+size 759654