Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77058.59005961688,
-      "required_pages_per_second_upper_estimate": 0.005331092060887144
+      "required_token_positions_per_second": 75822.40961744744,
+      "required_pages_per_second_upper_estimate": 0.005245570229564058
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9632.32375745211,
-      "required_pages_per_second_upper_estimate": 0.004032852811452092
+      "required_token_positions_per_second": 9477.80120218093,
+      "required_pages_per_second_upper_estimate": 0.003968157444357935
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19264.64751490422,
-      "required_pages_per_second_upper_estimate": 0.004012318034435174
+      "required_token_positions_per_second": 18955.60240436186,
+      "required_pages_per_second_upper_estimate": 0.003947952087976834
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28896.971272356328,
-      "required_pages_per_second_upper_estimate": 0.0017323293046574405
+      "required_token_positions_per_second": 28433.403606542794,
+      "required_pages_per_second_upper_estimate": 0.0017045391309187603
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28896.971272356328,
-      "required_pages_per_second_upper_estimate": 0.0022466124120818784
+      "required_token_positions_per_second": 28433.403606542794,
+      "required_pages_per_second_upper_estimate": 0.002210572065083548
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28896.971272356328,
-      "required_pages_per_second_upper_estimate": 0.3706593203314007
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28433.403606542794,
+      "required_pages_per_second_upper_estimate": 0.36471317205452464
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.24925434814855407,
-    "producer_p95_seconds": 0.28103425999870524,
-    "gpu_update_seconds": 0.9567784713288922,
-    "producer_batches_per_second": 4.01196611986085,
-    "gpu_batches_per_second": 1.0451740188207583,
-    "producer_headroom": 3.8385628111837713,
+    "producer_batch_seconds": 0.38157920347669005,
+    "producer_p95_seconds": 0.28838687199822743,
+    "gpu_update_seconds": 0.9723774326348301,
+    "producer_batches_per_second": 2.6206878962183486,
+    "gpu_batches_per_second": 1.028407248500535,
+    "producer_headroom": 2.548297768261972,
     "sustainable": true,
-    "buffer_seconds": 60.27704369372021,
+    "buffer_seconds": 60.28740082335947,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455745.37694572966,
-    "upload_bytes_per_second": 5897838.74281652,
+    "generation_bytes_per_second": 429427.70277391275,
+    "upload_bytes_per_second": 5894493.727171952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.79234868315618,
+    "estimated_full_shards_per_hour": 92.14518844998395,
     "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": 1.1649709810008062,
-      "bytes": 7159686,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.206743112001277,
       "bytes": 6257240,
@@ -562,6 +556,12 @@
       "bytes": 2427201,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.7313189029991918,
+      "bytes": 10259217,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 423900, "loss": 0.5853983163833618, "lr": 0.0003, "grad_norm": 0.15177953842616437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62047360879, "tokens_per_second": 151233.1366609056, "data_wait_s": 1.0456999007146806e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424000, "loss": 0.6058544516563416, "lr": 0.0003, "grad_norm": 0.17613054731147765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62062117848, "tokens_per_second": 153140.9930948525, "data_wait_s": 1.4223001926438883e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424100, "loss": 0.5639068484306335, "lr": 0.0003, "grad_norm": 0.14342668742039902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62076902751, "tokens_per_second": 152433.02420459464, "data_wait_s": 1.112700192607008e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 424200, "loss": 0.6424958109855652, "lr": 0.0003, "grad_norm": 0.14314316147947068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62091558430, "tokens_per_second": 151029.45185137034, "data_wait_s": 1.0495001333765686e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07fed0c43a630862f9d6080160d73b4f5ac8d53339a1c5ec8871464284c3239e
-size 575940
+oid sha256:9d1b117d24e3249857e9d1b2250158c728150d632b7b54bc2a49c73ccf6865ad
+size 578058