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": 75004.56578027054,
-      "required_pages_per_second_upper_estimate": 0.006501205664499835
+      "required_token_positions_per_second": 74915.65374802366,
+      "required_pages_per_second_upper_estimate": 0.00649349899489016
     },
     {
       "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": 9375.570722533817,
-      "required_pages_per_second_upper_estimate": 0.0037400553384928266
+      "required_token_positions_per_second": 9364.456718502957,
+      "required_pages_per_second_upper_estimate": 0.003735621796115748
     },
     {
       "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": 18751.141445067635,
-      "required_pages_per_second_upper_estimate": 0.004337746564498165
+      "required_token_positions_per_second": 18728.913437005915,
+      "required_pages_per_second_upper_estimate": 0.004332604506032667
     },
     {
       "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": 28126.71216760145,
-      "required_pages_per_second_upper_estimate": 0.0017127692015529532
+      "required_token_positions_per_second": 28093.370155508874,
+      "required_pages_per_second_upper_estimate": 0.0017107388479485333
     },
     {
       "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": 28126.71216760145,
-      "required_pages_per_second_upper_estimate": 0.0021652721623844837
+      "required_token_positions_per_second": 28093.370155508874,
+      "required_pages_per_second_upper_estimate": 0.0021627054019970004
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28126.71216760145,
-      "required_pages_per_second_upper_estimate": 0.349239631071451
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28093.370155508874,
+      "required_pages_per_second_upper_estimate": 0.3488256348697974
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10715751317178501,
-    "producer_p95_seconds": 0.3286345309970784,
-    "gpu_update_seconds": 0.9829801590477799,
-    "producer_batches_per_second": 9.332056804984756,
-    "gpu_batches_per_second": 1.0173145315249368,
-    "producer_headroom": 9.173226682406833,
+    "producer_batch_seconds": 0.08143923094544903,
+    "producer_p95_seconds": 0.28295052600151394,
+    "gpu_update_seconds": 0.984146787906059,
+    "producer_batches_per_second": 12.279094342993448,
+    "gpu_batches_per_second": 1.016108584907005,
+    "producer_headroom": 12.084431256052461,
     "sustainable": true,
-    "buffer_seconds": 60.94476986096235,
+    "buffer_seconds": 60.0329540622696,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458451.676564232,
-    "upload_bytes_per_second": 5079086.963043652,
+    "generation_bytes_per_second": 468931.9584054949,
+    "upload_bytes_per_second": 5076310.468121829,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.37305758185596,
+    "estimated_full_shards_per_hour": 100.621882096516,
     "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": 2.5150336740007333,
-      "bytes": 12238900,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 2.105870221999794,
       "bytes": 7597429,
@@ -562,6 +556,12 @@
       "bytes": 2665528,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3528436140004487,
+      "bytes": 6109175,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -302,3 +302,4 @@
 {"step": 385200, "loss": 0.540040971711278, "lr": 0.0003, "grad_norm": 0.14787405830661698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56382360865, "tokens_per_second": 147281.05100936638, "data_wait_s": 1.2210999557282776e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 385300, "loss": 0.5169158019125462, "lr": 0.0003, "grad_norm": 0.1593752303976673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56397003415, "tokens_per_second": 149602.2523016187, "data_wait_s": 9.520997991785407e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 385400, "loss": 0.5893670469522476, "lr": 0.0003, "grad_norm": 0.14939123712810912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56411634435, "tokens_per_second": 149417.05988998697, "data_wait_s": 1.6564998077228665e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 385500, "loss": 0.5525905806571245, "lr": 0.0003, "grad_norm": 0.15506915940897897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56426203371, "tokens_per_second": 148504.1289400462, "data_wait_s": 1.0281000868417323e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee79d3d9b93f517b942988bde0eb146aa01967def39c163471c924cfc04f1874
-size 648735
+oid sha256:fe91905380db0f10f3d83fce8d7b091d59b1af653eac37007b479fdb7a8df58d
+size 650853