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": 72239.82350495458,
-      "required_pages_per_second_upper_estimate": 0.006261564811250681
+      "required_token_positions_per_second": 74879.6943500283,
+      "required_pages_per_second_upper_estimate": 0.006490382125410183
     },
     {
       "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": 9029.977938119322,
-      "required_pages_per_second_upper_estimate": 0.003602193209717298
+      "required_token_positions_per_second": 9359.961793753537,
+      "required_pages_per_second_upper_estimate": 0.0037338287034280903
     },
     {
       "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": 18059.955876238644,
-      "required_pages_per_second_upper_estimate": 0.004177852947600122
+      "required_token_positions_per_second": 18719.923587507074,
+      "required_pages_per_second_upper_estimate": 0.004330524862567057
     },
     {
       "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": 27089.933814357966,
-      "required_pages_per_second_upper_estimate": 0.0016496348393960536
+      "required_token_positions_per_second": 28079.88538126061,
+      "required_pages_per_second_upper_estimate": 0.0017099176959459567
     },
     {
       "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": 27089.933814357966,
-      "required_pages_per_second_upper_estimate": 0.002085458094765631
+      "required_token_positions_per_second": 28079.88538126061,
+      "required_pages_per_second_upper_estimate": 0.0021616673067471246
     },
     {
       "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": 27089.933814357966,
-      "required_pages_per_second_upper_estimate": 0.3363663137981048
+      "required_token_positions_per_second": 28079.88538126061,
+      "required_pages_per_second_upper_estimate": 0.3486581991042702
     }
   ],
   "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": 59,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.17726908742986325,
-    "producer_p95_seconds": 0.5115464400005294,
-    "gpu_update_seconds": 1.0206005001513239,
-    "producer_batches_per_second": 5.641141467463419,
-    "gpu_batches_per_second": 0.9798153144660723,
-    "producer_headroom": 5.7573518031175395,
+    "producer_batch_seconds": 0.22502684622617153,
+    "producer_p95_seconds": 0.49266965000424534,
+    "gpu_update_seconds": 0.9846194036978216,
+    "producer_batches_per_second": 4.443914211884359,
+    "gpu_batches_per_second": 1.0156208543569376,
+    "producer_headroom": 4.375564161389853,
     "sustainable": true,
-    "buffer_seconds": 60.21542950892811,
+    "buffer_seconds": 60.06178362556712,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368167.85681965045,
-    "upload_bytes_per_second": 5463300.696845453,
+    "generation_bytes_per_second": 368963.9882737739,
+    "upload_bytes_per_second": 4378923.702938631,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.00025156442771,
+    "estimated_full_shards_per_hour": 79.17108284149087,
     "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": 1.216554687001917,
-      "bytes": 10409429,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1135386019959697,
       "bytes": 1273006,
@@ -562,6 +556,12 @@
       "bytes": 885141,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 20.492392911000934,
+      "bytes": 12112389,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 277900, "loss": 0.5784200374037027, "lr": 0.0003, "grad_norm": 0.15651964954651632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40676732679, "tokens_per_second": 146242.51833620656, "data_wait_s": 1.681399589870125e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278000, "loss": 0.5909800697118044, "lr": 0.0003, "grad_norm": 0.14002003126128926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40691426002, "tokens_per_second": 146438.251424613, "data_wait_s": 2.112299989676103e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278100, "loss": 0.5916734673082829, "lr": 0.0003, "grad_norm": 0.14871018837106634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40706113567, "tokens_per_second": 146715.18309275972, "data_wait_s": 1.917599729495123e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278200, "loss": 0.6078765690326691, "lr": 0.0003, "grad_norm": 0.1374363847299842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40720756769, "tokens_per_second": 146157.92697024057, "data_wait_s": 1.7257996660191566e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fddee14edf04abd3aca25d28a4f7a9d52fdebd8a3aaa3835eb73e3991e043fe
-size 874809
+oid sha256:b88941a55856a70d105b4d86efb6234f560bcec169c71a5f596c9e5c5a8c461d
+size 876927