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": 75649.66254309744,
-      "required_pages_per_second_upper_estimate": 0.0065571210169188725
+      "required_token_positions_per_second": 75987.22577544209,
+      "required_pages_per_second_upper_estimate": 0.006586380142352316
     },
     {
       "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": 9456.20781788718,
-      "required_pages_per_second_upper_estimate": 0.003772222681461297
+      "required_token_positions_per_second": 9498.40322193026,
+      "required_pages_per_second_upper_estimate": 0.003789055059011593
     },
     {
       "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": 18912.41563577436,
-      "required_pages_per_second_upper_estimate": 0.00437505451018935
+      "required_token_positions_per_second": 18996.80644386052,
+      "required_pages_per_second_upper_estimate": 0.004394576838412588
     },
     {
       "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": 28368.62345366154,
-      "required_pages_per_second_upper_estimate": 0.0017275003296635845
+      "required_token_positions_per_second": 28495.209665790782,
+      "required_pages_per_second_upper_estimate": 0.0017352087658357282
     },
     {
       "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": 28368.62345366154,
-      "required_pages_per_second_upper_estimate": 0.0021838951628386886
+      "required_token_positions_per_second": 28495.209665790782,
+      "required_pages_per_second_upper_estimate": 0.0021936401198614603
     },
     {
       "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": 28368.62345366154,
-      "required_pages_per_second_upper_estimate": 0.35224335961932457
+      "required_token_positions_per_second": 28495.209665790782,
+      "required_pages_per_second_upper_estimate": 0.3538151367171708
     }
   ],
   "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": 0.08047119468751873,
-    "producer_p95_seconds": 0.20676450099927024,
-    "gpu_update_seconds": 0.9745978702548386,
-    "producer_batches_per_second": 12.426806932383002,
-    "gpu_batches_per_second": 1.0260642163506055,
-    "producer_headroom": 12.111139570368538,
+    "producer_batch_seconds": 0.06468902746877347,
+    "producer_p95_seconds": 0.20195343700106605,
+    "gpu_update_seconds": 0.9702683477073033,
+    "producer_batches_per_second": 15.458572174125164,
+    "gpu_batches_per_second": 1.0306427107129188,
+    "producer_headroom": 14.998963281302517,
     "sustainable": true,
-    "buffer_seconds": 60.42506795579999,
+    "buffer_seconds": 60.156637557852804,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444721.90153672465,
-    "upload_bytes_per_second": 5933724.734321785,
+    "generation_bytes_per_second": 445107.41568412626,
+    "upload_bytes_per_second": 5922529.079591898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.42696747375778,
+    "estimated_full_shards_per_hour": 95.50968983548013,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 2547467,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.207064377000279,
+      "bytes": 6456430,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -116,3 +116,4 @@
 {"step": 126600, "loss": 0.6114125605672598, "lr": 0.0003, "grad_norm": 0.1337470624611966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18527324149, "tokens_per_second": 149839.31787375326, "data_wait_s": 1.622099989617709e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126700, "loss": 0.6048011872917414, "lr": 0.0003, "grad_norm": 0.14368404649454217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18542167820, "tokens_per_second": 151276.57830114447, "data_wait_s": 1.3267999747768044e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126800, "loss": 0.6097234869375825, "lr": 0.0003, "grad_norm": 0.14419275058677364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18556838342, "tokens_per_second": 149293.96007045152, "data_wait_s": 1.1334001101204194e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 126900, "loss": 0.6471667028963566, "lr": 0.0003, "grad_norm": 0.17106806992462145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18571312114, "tokens_per_second": 127553.6732559711, "data_wait_s": 1.4460998499998823e-05, "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:b8dbccca3a76079c848173bcab71d18894d600683e03119a4e9f0e52a8d154ed
-size 251655
+oid sha256:6e839eefaec0968f8a3b49bdca7e408457ee9e109c1ebc5b8c9efefdfeeb51ff
+size 253773