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": 77072.56531071175,
-      "required_pages_per_second_upper_estimate": 0.005332058901184938
+      "required_token_positions_per_second": 76972.51887933697,
+      "required_pages_per_second_upper_estimate": 0.005325137456922727
     },
     {
       "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": 9634.07066383897,
-      "required_pages_per_second_upper_estimate": 0.004033584204676716
+      "required_token_positions_per_second": 9621.564859917122,
+      "required_pages_per_second_upper_estimate": 0.0040283482857255215
     },
     {
       "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": 19268.14132767794,
-      "required_pages_per_second_upper_estimate": 0.004013045703497901
+      "required_token_positions_per_second": 19243.129719834244,
+      "required_pages_per_second_upper_estimate": 0.004007836445184514
     },
     {
       "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": 28902.211991516906,
-      "required_pages_per_second_upper_estimate": 0.0017326434777690024
+      "required_token_positions_per_second": 28864.69457975136,
+      "required_pages_per_second_upper_estimate": 0.0017303943662194288
     },
     {
       "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": 28902.211991516906,
-      "required_pages_per_second_upper_estimate": 0.0022470198549451263
+      "required_token_positions_per_second": 28864.69457975136,
+      "required_pages_per_second_upper_estimate": 0.002244103041202009
     },
     {
       "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": 28902.211991516906,
-      "required_pages_per_second_upper_estimate": 0.37072654264974675
+      "required_token_positions_per_second": 28864.69457975136,
+      "required_pages_per_second_upper_estimate": 0.3702453095746766
     }
   ],
   "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": 0.2299636946953001,
-    "producer_p95_seconds": 0.15779463699982443,
-    "gpu_update_seconds": 0.9566049826260692,
-    "producer_batches_per_second": 4.348512495961553,
-    "gpu_batches_per_second": 1.045363570295027,
-    "producer_headroom": 4.159808720648546,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.26611390544236,
+    "buffer_seconds": 60.344445883099446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338498.6403837515,
-    "upload_bytes_per_second": 5024201.155770956,
+    "generation_bytes_per_second": 337337.50420749735,
+    "upload_bytes_per_second": 5257801.55089001,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.63392838129433,
+    "estimated_full_shards_per_hour": 72.384775587737,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 6234900,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1657635070000651,
+      "bytes": 12844748,
+      "pages": 34,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -27,3 +27,4 @@
 {"step": 607100, "loss": 0.5816693305969238, "lr": 0.0001, "grad_norm": 0.1479278366716872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88861688463, "tokens_per_second": 150020.00815992078, "data_wait_s": 1.2137999874539673e-05, "gpu_peak_gb": 89.37861251831055}
 {"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}

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:a00eeabecb6cebc8d12344e9ff025c5f35339ca8b8f8fc420bf971f492b7ca2b
-size 62370
+oid sha256:551b28099d494722d36abbe7b5b9afc38089acdca9ce5f424cf1992293856d55
+size 64488