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": 73153.36449884849,
-      "required_pages_per_second_upper_estimate": 0.006340748229252924
+      "required_token_positions_per_second": 74329.22903120448,
+      "required_pages_per_second_upper_estimate": 0.00644266918671613
     },
     {
       "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": 9144.170562356061,
-      "required_pages_per_second_upper_estimate": 0.003647746354857213
+      "required_token_positions_per_second": 9291.15362890056,
+      "required_pages_per_second_upper_estimate": 0.003706380097694495
     },
     {
       "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": 18288.341124712122,
-      "required_pages_per_second_upper_estimate": 0.0042306858553913685
+      "required_token_positions_per_second": 18582.30725780112,
+      "required_pages_per_second_upper_estimate": 0.0042986897466542676
     },
     {
       "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": 27432.511687068185,
-      "required_pages_per_second_upper_estimate": 0.0016704960344769155
+      "required_token_positions_per_second": 27873.460886701683,
+      "required_pages_per_second_upper_estimate": 0.0016973475272529404
     },
     {
       "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": 27432.511687068185,
-      "required_pages_per_second_upper_estimate": 0.00211183068772311
+      "required_token_positions_per_second": 27873.460886701683,
+      "required_pages_per_second_upper_estimate": 0.0021457761777364053
     },
     {
       "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": 27432.511687068185,
-      "required_pages_per_second_upper_estimate": 0.34061998444278013
+      "required_token_positions_per_second": 27873.460886701683,
+      "required_pages_per_second_upper_estimate": 0.34609509774019004
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.20820850416403403,
-    "producer_p95_seconds": 0.1917496249998294,
-    "gpu_update_seconds": 1.007855216299184,
-    "producer_batches_per_second": 4.80287778837393,
-    "gpu_batches_per_second": 0.9922060072000934,
-    "producer_headroom": 4.840605432260154,
+    "producer_batch_seconds": 0.10062098585935786,
+    "producer_p95_seconds": 0.1909820259998014,
+    "gpu_update_seconds": 0.9919112704511965,
+    "producer_batches_per_second": 9.938284657613488,
+    "gpu_batches_per_second": 1.0081546906359116,
+    "producer_headroom": 9.85789656083903,
     "sustainable": true,
-    "buffer_seconds": 60.471312977951044,
+    "buffer_seconds": 60.506587497522986,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393512.74205682176,
-    "upload_bytes_per_second": 5377575.293997336,
+    "generation_bytes_per_second": 459625.16079449747,
+    "upload_bytes_per_second": 5452156.9477346465,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.43867393759241,
+    "estimated_full_shards_per_hour": 98.62485998035615,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 6061414,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.345420446999924,
+      "bytes": 12056021,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -64,3 +64,4 @@
 {"step": 321400, "loss": 0.5764908324927092, "lr": 0.0003, "grad_norm": 0.1483505579466562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47043747942, "tokens_per_second": 147907.25459283905, "data_wait_s": 1.2944000445713755e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 321500, "loss": 0.5943129621446133, "lr": 0.0003, "grad_norm": 0.17142695421148932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47058330587, "tokens_per_second": 148283.6343889757, "data_wait_s": 8.058000275923405e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 321600, "loss": 0.6203801073133945, "lr": 0.0003, "grad_norm": 0.15273625784521466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47072973219, "tokens_per_second": 149159.19618488895, "data_wait_s": 2.0823000340897124e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 321700, "loss": 0.5895306020975113, "lr": 0.0003, "grad_norm": 0.14704612855299368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47087634270, "tokens_per_second": 149354.38844349168, "data_wait_s": 1.557199993840186e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87bec35287ad1320a0dddc7020845049c93c21ebc54de2d656477a63279b09c6
-size 140736
+oid sha256:0349c30a874f1d71dae5ab938c72861c4af06d6dcee9960bb239ed285d5d13fc
+size 142854