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": 74321.694413776,
-      "required_pages_per_second_upper_estimate": 0.006442016105173743
+      "required_token_positions_per_second": 73547.07422008907,
+      "required_pages_per_second_upper_estimate": 0.006374873990041892
     },
     {
       "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": 9290.211801722,
-      "required_pages_per_second_upper_estimate": 0.0037060043887513962
+      "required_token_positions_per_second": 9193.384277511133,
+      "required_pages_per_second_upper_estimate": 0.003667378441643184
     },
     {
       "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": 18580.423603444,
-      "required_pages_per_second_upper_estimate": 0.004298253996369932
+      "required_token_positions_per_second": 18386.768555022267,
+      "required_pages_per_second_upper_estimate": 0.004253455309130011
     },
     {
       "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": 27870.635405166002,
-      "required_pages_per_second_upper_estimate": 0.0016971754702515733
+      "required_token_positions_per_second": 27580.1528325334,
+      "required_pages_per_second_upper_estimate": 0.0016794866056225222
     },
     {
       "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": 27870.635405166002,
-      "required_pages_per_second_upper_estimate": 0.002145558664346355
+      "required_token_positions_per_second": 27580.1528325334,
+      "required_pages_per_second_upper_estimate": 0.002123196511797871
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27870.635405166002,
-      "required_pages_per_second_upper_estimate": 0.34606001471579523
+      "required_token_positions_per_second": 27580.1528325334,
+      "required_pages_per_second_upper_estimate": 0.3424531933463303
     }
   ],
   "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": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11464767338281945,
-    "producer_p95_seconds": 0.2934116480028024,
-    "gpu_update_seconds": 0.9920118288682885,
-    "producer_batches_per_second": 8.722374998931773,
-    "gpu_batches_per_second": 1.0080524958465713,
-    "producer_headroom": 8.652699174765344,
+    "producer_batch_seconds": 0.3825688774765297,
+    "producer_p95_seconds": 0.308603129000403,
+    "gpu_update_seconds": 1.0024599996917554,
+    "producer_batches_per_second": 2.613908393688792,
+    "gpu_batches_per_second": 0.9975460370563296,
+    "producer_headroom": 2.620338607531543,
     "sustainable": true,
-    "buffer_seconds": 60.5127215609656,
+    "buffer_seconds": 60.14759998150532,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378593.23161775264,
-    "upload_bytes_per_second": 5822134.6441319715,
+    "generation_bytes_per_second": 364698.6120124421,
+    "upload_bytes_per_second": 5741572.762275082,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.237294305796,
+    "estimated_full_shards_per_hour": 78.25583238868663,
     "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": 1.3716071430008014,
-      "bytes": 11837179,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.8982555259990477,
       "bytes": 1117404,
@@ -562,6 +556,12 @@
       "bytes": 4032617,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.6352041230020404,
+      "bytes": 7545726,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -201,3 +201,4 @@
 {"step": 375100, "loss": 0.6094298958778381, "lr": 0.0003, "grad_norm": 0.16543400737552919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54903847811, "tokens_per_second": 125171.26088617968, "data_wait_s": 1.2852997315349057e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375200, "loss": 0.5757131744176149, "lr": 0.0003, "grad_norm": 0.15580247337755893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54918289514, "tokens_per_second": 147168.59554686316, "data_wait_s": 1.3649998436449096e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375300, "loss": 0.5982198733836412, "lr": 0.0003, "grad_norm": 0.16705821078006447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54932780320, "tokens_per_second": 143978.49520058522, "data_wait_s": 7.730999641353264e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 375400, "loss": 0.5743808448314667, "lr": 0.0003, "grad_norm": 0.15233685722282306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54947279592, "tokens_per_second": 147841.1089620428, "data_wait_s": 9.740000677993521e-06, "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:d3671a6378968ac5a4bf8ed23bc3b2b99bb5040f0fa69a36382582f989ec8ea1
-size 433251
+oid sha256:a72650ed01e5fa2e34ff730a8f95e8f16f9bb74d75b181ee5e118972e716d61f
+size 435369