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": 77115.79531992723,
-      "required_pages_per_second_upper_estimate": 0.005335049653529905
+      "required_token_positions_per_second": 76978.42092638944,
+      "required_pages_per_second_upper_estimate": 0.0053255457742324526
     },
     {
       "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": 9639.474414990904,
-      "required_pages_per_second_upper_estimate": 0.004035846642440876
+      "required_token_positions_per_second": 9622.30261579868,
+      "required_pages_per_second_upper_estimate": 0.004028657168707035
     },
     {
       "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": 19278.948829981808,
-      "required_pages_per_second_upper_estimate": 0.004015296621214795
+      "required_token_positions_per_second": 19244.60523159736,
+      "required_pages_per_second_upper_estimate": 0.00400814375537291
     },
     {
       "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": 28918.423244972713,
-      "required_pages_per_second_upper_estimate": 0.0017336153176605272
+      "required_token_positions_per_second": 28866.907847396044,
+      "required_pages_per_second_upper_estimate": 0.0017305270482352634
     },
     {
       "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": 28918.423244972713,
-      "required_pages_per_second_upper_estimate": 0.002248280208595547
+      "required_token_positions_per_second": 28866.907847396044,
+      "required_pages_per_second_upper_estimate": 0.0022442751130262477
     },
     {
       "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": 28918.423244972713,
-      "required_pages_per_second_upper_estimate": 0.37093448320279004
+      "required_token_positions_per_second": 28866.907847396044,
+      "required_pages_per_second_upper_estimate": 0.3702736989956009
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.343836432000046,
-    "producer_p95_seconds": 0.22972871499950998,
-    "gpu_update_seconds": 0.9560687235880481,
-    "producer_batches_per_second": 2.908359635374143,
-    "gpu_batches_per_second": 1.0459499148210616,
-    "producer_headroom": 2.7805916843271574,
+    "producer_batch_seconds": 0.0596970721484098,
+    "producer_p95_seconds": 0.1938023249986145,
+    "gpu_update_seconds": 0.9577749077303409,
+    "producer_batches_per_second": 16.751240287194516,
+    "gpu_batches_per_second": 1.0440866553601,
+    "producer_headroom": 16.0439176204365,
     "sustainable": true,
-    "buffer_seconds": 60.23232958604703,
+    "buffer_seconds": 60.33981918701148,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 599114.8376553796,
-    "upload_bytes_per_second": 5275140.440368012,
+    "generation_bytes_per_second": 601520.1839787205,
+    "upload_bytes_per_second": 5389721.567913899,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.2780487406065,
+    "estimated_full_shards_per_hour": 64.53611440430265,
     "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": 1.609872815000017,
-      "bytes": 7812095,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9189778770000885,
       "bytes": 1227062,
@@ -562,6 +556,12 @@
       "bytes": 2736110,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.411381372001415,
+      "bytes": 15991084,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -104,3 +104,4 @@
 {"step": 895400, "loss": 0.5063887238502502, "lr": 3e-05, "grad_norm": 0.325257508165069, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131061408234, "tokens_per_second": 153339.8873856963, "data_wait_s": 1.215900010720361e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895500, "loss": 0.5193361639976501, "lr": 3e-05, "grad_norm": 0.3058153376724782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131076165805, "tokens_per_second": 153918.2519494799, "data_wait_s": 2.2959000489208847e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895600, "loss": 0.5518725514411926, "lr": 3e-05, "grad_norm": 0.3277005091052649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131090877456, "tokens_per_second": 153869.18445872568, "data_wait_s": 1.3012999261263758e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 895700, "loss": 0.5045432448387146, "lr": 3e-05, "grad_norm": 0.3197829840439839, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131105489841, "tokens_per_second": 151838.63309223627, "data_wait_s": 1.6257999959634617e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa0de1bfa0bc1a2f6f83689c7cd3b9cde2cffcaa01008a84d4966c9f25ff17e9
-size 226239
+oid sha256:6e5ce78e0d864f9308c65d725e94e40db374abb5dc7e28767bc004ab5f35ac93
+size 228357