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": 74502.39919448904,
-      "required_pages_per_second_upper_estimate": 0.006457679137573871
+      "required_token_positions_per_second": 74393.59217399382,
+      "required_pages_per_second_upper_estimate": 0.006448248020806216
     },
     {
       "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": 9312.79989931113,
-      "required_pages_per_second_upper_estimate": 0.0037150151186018552
+      "required_token_positions_per_second": 9299.199021749228,
+      "required_pages_per_second_upper_estimate": 0.003709589525191171
     },
     {
       "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": 18625.59979862226,
-      "required_pages_per_second_upper_estimate": 0.004308704714050543
+      "required_token_positions_per_second": 18598.398043498455,
+      "required_pages_per_second_upper_estimate": 0.00430241206673719
     },
     {
       "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": 27938.399697933393,
-      "required_pages_per_second_upper_estimate": 0.0017013019601493404
+      "required_token_positions_per_second": 27897.59706524768,
+      "required_pages_per_second_upper_estimate": 0.0016988172938936495
     },
     {
       "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": 27938.399697933393,
-      "required_pages_per_second_upper_estimate": 0.002150775347187157
+      "required_token_positions_per_second": 27897.59706524768,
+      "required_pages_per_second_upper_estimate": 0.0021476342475740943
     },
     {
       "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": 27938.399697933393,
-      "required_pages_per_second_upper_estimate": 0.3469014204394675
+      "required_token_positions_per_second": 27897.59706524768,
+      "required_pages_per_second_upper_estimate": 0.346394788299138
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0719678307421816,
-    "producer_p95_seconds": 0.19274514999960957,
-    "gpu_update_seconds": 0.989605714676819,
-    "producer_batches_per_second": 13.89509715225976,
-    "gpu_batches_per_second": 1.0105034612967807,
-    "producer_headroom": 13.750667547865852,
+    "producer_batch_seconds": 0.08422711268749339,
+    "producer_p95_seconds": 0.18170365799960564,
+    "gpu_update_seconds": 0.9910530980620332,
+    "producer_batches_per_second": 11.872661523021515,
+    "gpu_batches_per_second": 1.0090276716307756,
+    "producer_headroom": 11.76643798463237,
     "sustainable": true,
-    "buffer_seconds": 60.36594859528596,
+    "buffer_seconds": 60.45423898178403,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 407270.7114728785,
-    "upload_bytes_per_second": 5363445.222218913,
+    "generation_bytes_per_second": 403893.9982961527,
+    "upload_bytes_per_second": 5353668.362590939,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.39081390514151,
+    "estimated_full_shards_per_hour": 86.66624986327587,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 1180635,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.2576917689993934,
+      "bytes": 6183813,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -57,3 +57,4 @@
 {"step": 320700, "loss": 0.5956653617322445, "lr": 0.0003, "grad_norm": 0.1756550220108485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46941285266, "tokens_per_second": 146610.39086905512, "data_wait_s": 1.580099979037186e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320800, "loss": 0.5841874033212662, "lr": 0.0003, "grad_norm": 0.1514964677834582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46955936745, "tokens_per_second": 148588.1178477004, "data_wait_s": 1.2708999747701455e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320900, "loss": 0.5831205900758505, "lr": 0.0003, "grad_norm": 0.15360701551341333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46970679323, "tokens_per_second": 150797.79150006865, "data_wait_s": 1.4571000065188855e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 321000, "loss": 0.6045909468084574, "lr": 0.0003, "grad_norm": 0.1401154045649198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46985391259, "tokens_per_second": 149704.20951458134, "data_wait_s": 9.342999874206726e-06, "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 sha256:2dd37ff1f8b4f7be3de8f410eb783f10b5cbab5f4eb23c82b1801dd5ba951b03
-size 125910
+oid sha256:5cfe51addf86c67b0a2c3be1f2db1c2696e50574664a568063ab0a3fd28e01af
+size 128028