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": 77162.19756429653,
-      "required_pages_per_second_upper_estimate": 0.005338259868463414
+      "required_token_positions_per_second": 77415.48812361607,
+      "required_pages_per_second_upper_estimate": 0.005355783097072209
     },
     {
       "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": 9645.274695537066,
-      "required_pages_per_second_upper_estimate": 0.004038275098781923
+      "required_token_positions_per_second": 9676.936015452009,
+      "required_pages_per_second_upper_estimate": 0.004051531032266766
     },
     {
       "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": 19290.54939107413,
-      "required_pages_per_second_upper_estimate": 0.004017712712162958
+      "required_token_positions_per_second": 19353.872030904018,
+      "required_pages_per_second_upper_estimate": 0.004030901148109212
     },
     {
       "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": 28935.824086611196,
-      "required_pages_per_second_upper_estimate": 0.0017346584715472076
+      "required_token_positions_per_second": 29030.808046356025,
+      "required_pages_per_second_upper_estimate": 0.001740352614901801
     },
     {
       "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": 28935.824086611196,
-      "required_pages_per_second_upper_estimate": 0.002249633047494725
+      "required_token_positions_per_second": 29030.808046356025,
+      "required_pages_per_second_upper_estimate": 0.0022570176325746004
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28935.824086611196,
-      "required_pages_per_second_upper_estimate": 0.37115768251576037
+      "required_token_positions_per_second": 29030.808046356025,
+      "required_pages_per_second_upper_estimate": 0.37237603476553693
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08996185067968554,
-    "producer_p95_seconds": 0.2137298779998673,
-    "gpu_update_seconds": 0.9554937822832881,
-    "producer_batches_per_second": 11.115822900982316,
-    "gpu_batches_per_second": 1.0465792855400462,
-    "producer_headroom": 10.621099666850784,
+    "producer_batch_seconds": 0.4394334593517044,
+    "producer_p95_seconds": 0.21852520599713898,
+    "gpu_update_seconds": 0.9523675660647138,
+    "producer_batches_per_second": 2.275657391850176,
+    "gpu_batches_per_second": 1.0500147586210948,
+    "producer_headroom": 2.167262291473527,
     "sustainable": true,
-    "buffer_seconds": 60.196108283847146,
+    "buffer_seconds": 60.951524228141686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316455.49775540904,
-    "upload_bytes_per_second": 4722495.633900892,
+    "generation_bytes_per_second": 296988.82874613756,
+    "upload_bytes_per_second": 4767584.704755423,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.90398311135009,
+    "estimated_full_shards_per_hour": 63.726889102822255,
     "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": 1.253272633000961,
-      "bytes": 6708450,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 1.5571684259994072,
       "bytes": 5848121,
@@ -562,6 +556,12 @@
       "bytes": 7489532,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0948180369996408,
+      "bytes": 10021223,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -197,3 +197,4 @@
 {"step": 944700, "loss": 0.5725507736206055, "lr": 3e-05, "grad_norm": 0.34738436125650035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138277526113, "tokens_per_second": 152562.25768287326, "data_wait_s": 1.4356002793647349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944800, "loss": 0.5285707116127014, "lr": 3e-05, "grad_norm": 0.3409574839623096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138292218106, "tokens_per_second": 153583.49825800018, "data_wait_s": 7.965001714183018e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 944900, "loss": 0.5289706587791443, "lr": 3e-05, "grad_norm": 0.35097219719823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138306891113, "tokens_per_second": 152803.21131051768, "data_wait_s": 1.2256998161319643e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 945000, "loss": 0.547132670879364, "lr": 3e-05, "grad_norm": 0.3489687738519755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138321369205, "tokens_per_second": 150205.0161449377, "data_wait_s": 1.516900010756217e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09840f8bc083df2ce5e75f0f94d4b3b98905d3ae424db743c61ca14f61468120
-size 423996
+oid sha256:0aa57f951eeb501061b23f747c4c4d89cb1c16c2a55ea5edabdc53761edee676
+size 426114