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": 76972.51887933697,
-      "required_pages_per_second_upper_estimate": 0.005325137456922727
+      "required_token_positions_per_second": 76985.36759228885,
+      "required_pages_per_second_upper_estimate": 0.00532602636069786
     },
     {
       "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": 9621.564859917122,
-      "required_pages_per_second_upper_estimate": 0.0040283482857255215
+      "required_token_positions_per_second": 9623.170949036106,
+      "required_pages_per_second_upper_estimate": 0.004029020721700686
     },
     {
       "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": 19243.129719834244,
-      "required_pages_per_second_upper_estimate": 0.004007836445184514
+      "required_token_positions_per_second": 19246.341898072213,
+      "required_pages_per_second_upper_estimate": 0.004008505457200647
     },
     {
       "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": 28864.69457975136,
-      "required_pages_per_second_upper_estimate": 0.0017303943662194288
+      "required_token_positions_per_second": 28869.512847108323,
+      "required_pages_per_second_upper_estimate": 0.0017306832139904102
     },
     {
       "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": 28864.69457975136,
-      "required_pages_per_second_upper_estimate": 0.002244103041202009
+      "required_token_positions_per_second": 28869.512847108323,
+      "required_pages_per_second_upper_estimate": 0.002244477640295695
     },
     {
       "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": 28864.69457975136,
-      "required_pages_per_second_upper_estimate": 0.3702453095746766
+      "required_token_positions_per_second": 28869.512847108323,
+      "required_pages_per_second_upper_estimate": 0.3703071131348793
     }
   ],
   "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": 0.06533507758591739,
-    "producer_p95_seconds": 0.14802258500003518,
-    "gpu_update_seconds": 0.9578483473507848,
-    "producer_batches_per_second": 15.305713820955873,
-    "gpu_batches_per_second": 1.044006603723646,
-    "producer_headroom": 14.660552688426648,
+    "producer_batch_seconds": 0.06568073322656431,
+    "producer_p95_seconds": 0.15397351499996148,
+    "gpu_update_seconds": 0.9576884842644419,
+    "producer_batches_per_second": 15.225164989412054,
+    "gpu_batches_per_second": 1.0441808755464526,
+    "producer_headroom": 14.580965181386077,
     "sustainable": true,
-    "buffer_seconds": 60.344445883099446,
+    "buffer_seconds": 60.334374508659835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337337.50420749735,
-    "upload_bytes_per_second": 5257801.55089001,
+    "generation_bytes_per_second": 336490.4636753246,
+    "upload_bytes_per_second": 5275125.526922865,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.384775587737,
+    "estimated_full_shards_per_hour": 72.20302040762714,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 12844748,
       "pages": 34,
       "shards": 5
+    },
+    {
+      "seconds": 1.0483822090000103,
+      "bytes": 6048562,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 607200, "loss": 0.5858821272850037, "lr": 0.0001, "grad_norm": 0.17613203362462435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88876154434, "tokens_per_second": 150519.37055730794, "data_wait_s": 1.4258999726735055e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 607300, "loss": 0.5235000848770142, "lr": 0.0001, "grad_norm": 0.14691208586594298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88890668412, "tokens_per_second": 150220.73112838183, "data_wait_s": 1.115099985327106e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 607400, "loss": 0.5584568977355957, "lr": 0.0001, "grad_norm": 0.13680674459670614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88905380803, "tokens_per_second": 152331.02339301273, "data_wait_s": 1.1147999884997262e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 607500, "loss": 0.5816423296928406, "lr": 0.0001, "grad_norm": 0.16900962176126952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88920076367, "tokens_per_second": 152590.53688233954, "data_wait_s": 1.2109000181226293e-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 sha256:551b28099d494722d36abbe7b5b9afc38089acdca9ce5f424cf1992293856d55
-size 64488
+oid sha256:00e26f574e9c578c987a5ef673f67b3c2e307c751920bf8460c9c7d278dae3c7
+size 66606