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": 73412.33330635568,
-      "required_pages_per_second_upper_estimate": 0.006363194989137205
+      "required_token_positions_per_second": 74962.50138789843,
+      "required_pages_per_second_upper_estimate": 0.006497559629580248
     },
     {
       "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": 9176.54166329446,
-      "required_pages_per_second_upper_estimate": 0.0036606596710126297
+      "required_token_positions_per_second": 9370.312673487304,
+      "required_pages_per_second_upper_estimate": 0.003737957824113333
     },
     {
       "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": 18353.08332658892,
-      "required_pages_per_second_upper_estimate": 0.004245662824371729
+      "required_token_positions_per_second": 18740.625346974608,
+      "required_pages_per_second_upper_estimate": 0.004335313850281348
     },
     {
       "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": 27529.624989883378,
-      "required_pages_per_second_upper_estimate": 0.0016764097251042382
+      "required_token_positions_per_second": 28110.938020461912,
+      "required_pages_per_second_upper_estimate": 0.0017118086387527094
     },
     {
       "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": 27529.624989883378,
-      "required_pages_per_second_upper_estimate": 0.0021193067386006516
+      "required_token_positions_per_second": 28110.938020461912,
+      "required_pages_per_second_upper_estimate": 0.002164057824872048
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27529.624989883378,
-      "required_pages_per_second_upper_estimate": 0.3418258066464281
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28110.938020461912,
+      "required_pages_per_second_upper_estimate": 0.34904376895665234
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.28797938549996616,
-    "producer_p95_seconds": 0.17813776599996345,
-    "gpu_update_seconds": 1.0042999136443058,
-    "producer_batches_per_second": 3.472470775169834,
-    "gpu_batches_per_second": 0.9957184964512217,
-    "producer_headroom": 3.4874020996354402,
+    "producer_batch_seconds": 0.05080553835940549,
+    "producer_p95_seconds": 0.17008729799999855,
+    "gpu_update_seconds": 0.9835317476732743,
+    "producer_batches_per_second": 19.682893485467275,
+    "gpu_batches_per_second": 1.0167439966891607,
+    "producer_headroom": 19.358750629028535,
     "sustainable": true,
-    "buffer_seconds": 60.25799481865835,
+    "buffer_seconds": 60.97896835574301,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382965.7129730467,
-    "upload_bytes_per_second": 5369362.10550369,
+    "generation_bytes_per_second": 383249.0272204962,
+    "upload_bytes_per_second": 5393018.259402642,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.17552701848555,
+    "estimated_full_shards_per_hour": 82.23631966077008,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 6144064,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1848053769999751,
+      "bytes": 6754320,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 199400, "loss": 0.5781487785279751, "lr": 0.0003, "grad_norm": 0.1592378748404679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29184147161, "tokens_per_second": 147649.23572227385, "data_wait_s": 1.5172000530583318e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199500, "loss": 0.6128964312374592, "lr": 0.0003, "grad_norm": 0.1499064107232924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29198851808, "tokens_per_second": 147593.2937842481, "data_wait_s": 1.0265000128129032e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199600, "loss": 0.5836126524955034, "lr": 0.0003, "grad_norm": 0.14907554180658394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29213631440, "tokens_per_second": 149787.08687137134, "data_wait_s": 2.0182999833195936e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 199700, "loss": 0.5929410867393017, "lr": 0.0003, "grad_norm": 0.14810363831661824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29228097396, "tokens_per_second": 145247.3200887589, "data_wait_s": 1.8230999557999894e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f89052124ab099eac5681cf448fe9f44085555e4d52c1624c6603453c0813f3d
-size 97593
+oid sha256:a4548f9cc5ccdc9ba53accdf7272d9dcbcbb868ae1e42514a5ed377ae3dc1ae9
+size 99711