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": 76699.73446139753,
-      "required_pages_per_second_upper_estimate": 0.0053062655979426235
+      "required_token_positions_per_second": 76441.95494032078,
+      "required_pages_per_second_upper_estimate": 0.005288431812543639
     },
     {
       "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": 9587.466807674691,
-      "required_pages_per_second_upper_estimate": 0.004014072143299917
+      "required_token_positions_per_second": 9555.244367540097,
+      "required_pages_per_second_upper_estimate": 0.004000581280496627
     },
     {
       "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": 19174.933615349382,
-      "required_pages_per_second_upper_estimate": 0.0039936329950725335
+      "required_token_positions_per_second": 19110.488735080195,
+      "required_pages_per_second_upper_estimate": 0.003980210826038243
     },
     {
       "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": 28762.400423024075,
-      "required_pages_per_second_upper_estimate": 0.0017242619877177612
+      "required_token_positions_per_second": 28665.73310262029,
+      "required_pages_per_second_upper_estimate": 0.0017184669294620061
     },
     {
       "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": 28762.400423024075,
-      "required_pages_per_second_upper_estimate": 0.0022361501204609066
+      "required_token_positions_per_second": 28665.73310262029,
+      "required_pages_per_second_upper_estimate": 0.0022286346614941188
     },
     {
       "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": 28762.400423024075,
-      "required_pages_per_second_upper_estimate": 0.36893318996708707
+      "required_token_positions_per_second": 28665.73310262029,
+      "required_pages_per_second_upper_estimate": 0.36769324537422926
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10674078339039283,
-    "producer_p95_seconds": 0.26070677200186765,
-    "gpu_update_seconds": 0.9612549576310048,
-    "producer_batches_per_second": 9.368490358016285,
-    "gpu_batches_per_second": 1.040306728263313,
-    "producer_headroom": 9.00550780216142,
+    "producer_batch_seconds": 0.09151863030444929,
+    "producer_p95_seconds": 0.26772537600118085,
+    "gpu_update_seconds": 0.9644965262539453,
+    "producer_batches_per_second": 10.926736956982012,
+    "gpu_batches_per_second": 1.0368103697417639,
+    "producer_headroom": 10.538799838299756,
     "sustainable": true,
-    "buffer_seconds": 60.5590623307533,
+    "buffer_seconds": 60.763281153998555,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359982.97748261393,
-    "upload_bytes_per_second": 3895265.6047551525,
+    "generation_bytes_per_second": 359409.9930187503,
+    "upload_bytes_per_second": 3939915.8061535168,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.24396699293912,
+    "estimated_full_shards_per_hour": 77.12101786538965,
     "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": [
-    {
-      "seconds": 1.4790617950002343,
-      "bytes": 1048131,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.3689376410002296,
       "bytes": 5834070,
@@ -562,6 +556,12 @@
       "bytes": 2415576,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.5388405060002697,
+      "bytes": 6065532,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 822300, "loss": 0.5214368104934692, "lr": 3e-05, "grad_norm": 0.28538880079113754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120361651628, "tokens_per_second": 152933.67619670348, "data_wait_s": 1.2795997463399544e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822400, "loss": 0.524349570274353, "lr": 3e-05, "grad_norm": 0.27834077863244744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120376319050, "tokens_per_second": 151681.88975819264, "data_wait_s": 1.0267001925967634e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822500, "loss": 0.5585592985153198, "lr": 3e-05, "grad_norm": 0.2840605367832263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120390929173, "tokens_per_second": 150758.69757127998, "data_wait_s": 1.357400105916895e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 822600, "loss": 0.47821998596191406, "lr": 3e-05, "grad_norm": 0.26762885928740976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120405455371, "tokens_per_second": 150056.61543589283, "data_wait_s": 1.017099930322729e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98745b10738b5fd8f53e0cf12a3e1a703d70fcd9f3414718ae624fb631f9740f
-size 479847
+oid sha256:16dce1ff65a6052ab0f1f3d27a5ce15729e8cda5cd86b2a48755710599bbc05e
+size 481965