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": 77571.26574998416,
-      "required_pages_per_second_upper_estimate": 0.005366560154718245
+      "required_token_positions_per_second": 77186.22582919791,
+      "required_pages_per_second_upper_estimate": 0.005339922199582548
     },
     {
       "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": 9696.40821874802,
-      "required_pages_per_second_upper_estimate": 0.004059683637160962
+      "required_token_positions_per_second": 9648.278228649739,
+      "required_pages_per_second_upper_estimate": 0.004039532615375295
     },
     {
       "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": 19392.81643749604,
-      "required_pages_per_second_upper_estimate": 0.0040390122409692635
+      "required_token_positions_per_second": 19296.556457299477,
+      "required_pages_per_second_upper_estimate": 0.004018963825640707
     },
     {
       "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": 29089.22465624406,
-      "required_pages_per_second_upper_estimate": 0.0017438546014675901
+      "required_token_positions_per_second": 28944.834685949216,
+      "required_pages_per_second_upper_estimate": 0.0017351986432191332
     },
     {
       "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": 29089.22465624406,
-      "required_pages_per_second_upper_estimate": 0.0022615592670458835
+      "required_token_positions_per_second": 28944.834685949216,
+      "required_pages_per_second_upper_estimate": 0.0022503335819598186
     },
     {
       "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": 29089.22465624406,
-      "required_pages_per_second_upper_estimate": 0.37312534031431177
+      "required_token_positions_per_second": 28944.834685949216,
+      "required_pages_per_second_upper_estimate": 0.3712732608092407
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05626329523441598,
-    "producer_p95_seconds": 0.1819149599996308,
-    "gpu_update_seconds": 0.9504550336670903,
-    "producer_batches_per_second": 17.773576820084738,
-    "gpu_batches_per_second": 1.052127627902346,
-    "producer_headroom": 16.892985554918255,
+    "producer_batch_seconds": 0.5090261572579067,
+    "producer_p95_seconds": 0.18384084399986023,
+    "gpu_update_seconds": 0.9551963346821689,
+    "producer_batches_per_second": 1.964535585728914,
+    "gpu_batches_per_second": 1.04690518974064,
+    "producer_headroom": 1.8765171908409466,
     "sustainable": true,
-    "buffer_seconds": 60.82912215469378,
+    "buffer_seconds": 60.17736908497664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324038.0284101658,
-    "upload_bytes_per_second": 4252787.429657175,
+    "generation_bytes_per_second": 298927.24372286384,
+    "upload_bytes_per_second": 4263491.428027692,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.53101767758112,
+    "estimated_full_shards_per_hour": 64.1428278328365,
     "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": [
@@ -460,6 +460,18 @@
       "bytes": 6201672,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2297464679995755,
+      "bytes": 644185,
+      "pages": 19,
+      "shards": 1
+    },
+    {
+      "seconds": 1.3486207620007917,
+      "bytes": 11139901,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -98,3 +98,4 @@
 {"step": 934800, "loss": 0.5276320576667786, "lr": 3e-05, "grad_norm": 0.34994359220643617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136828241613, "tokens_per_second": 153607.4501460343, "data_wait_s": 1.6230998880928382e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 934900, "loss": 0.5477813482284546, "lr": 3e-05, "grad_norm": 0.3481558230709863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136843065163, "tokens_per_second": 154372.3115513618, "data_wait_s": 1.0394000128144398e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935000, "loss": 0.5534255504608154, "lr": 3e-05, "grad_norm": 0.35179548294994867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136857750792, "tokens_per_second": 132722.11465712264, "data_wait_s": 1.3328000932233408e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 935100, "loss": 0.528160035610199, "lr": 3e-05, "grad_norm": 0.3339950495529836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136872314892, "tokens_per_second": 140259.65118426422, "data_wait_s": 1.6283998775179498e-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 sha256:808e24ddd92a85e0548a9407bee3b7bda0f7fb465db6a9c89dbdf6fea41a38a3
-size 213531
+oid sha256:65cd4d0406cb8991c9a758de96e04caf647d0fd2724a4c0ad5b23f898af20ce7
+size 215649