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": 77448.39442336257,
-      "required_pages_per_second_upper_estimate": 0.005358059631241806
+      "required_token_positions_per_second": 77072.80063236308,
+      "required_pages_per_second_upper_estimate": 0.005332075181282798
     },
     {
       "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": 9681.04930292032,
-      "required_pages_per_second_upper_estimate": 0.0040532531798345385
+      "required_token_positions_per_second": 9634.100079045385,
+      "required_pages_per_second_upper_estimate": 0.004033596520209384
     },
     {
       "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": 19362.09860584064,
-      "required_pages_per_second_upper_estimate": 0.004032614526719141
+      "required_token_positions_per_second": 19268.20015809077,
+      "required_pages_per_second_upper_estimate": 0.004013057956321432
     },
     {
       "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": 29043.14790876096,
-      "required_pages_per_second_upper_estimate": 0.001741092370811098
+      "required_token_positions_per_second": 28902.300237136154,
+      "required_pages_per_second_upper_estimate": 0.001732648767959133
     },
     {
       "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": 29043.14790876096,
-      "required_pages_per_second_upper_estimate": 0.0022579770025992654
+      "required_token_positions_per_second": 28902.300237136154,
+      "required_pages_per_second_upper_estimate": 0.0022470267156538713
     },
     {
       "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": 29043.14790876096,
-      "required_pages_per_second_upper_estimate": 0.3725343172709555
+      "required_token_positions_per_second": 28902.300237136154,
+      "required_pages_per_second_upper_estimate": 0.3707276745697997
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0717435841249312,
-    "producer_p95_seconds": 0.2592391209982452,
-    "gpu_update_seconds": 0.9519629238144633,
-    "producer_batches_per_second": 13.938528611264289,
-    "gpu_batches_per_second": 1.0504610788759028,
-    "producer_headroom": 13.268962450450704,
+    "producer_batch_seconds": 0.09907552693766775,
+    "producer_p95_seconds": 0.26541559499673895,
+    "gpu_update_seconds": 0.9566020618827936,
+    "producer_batches_per_second": 10.093309931413625,
+    "gpu_batches_per_second": 1.0453667620491953,
+    "producer_headroom": 9.655281091612352,
     "sustainable": true,
-    "buffer_seconds": 60.925627124125654,
+    "buffer_seconds": 60.265929898616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263838.9817945861,
-    "upload_bytes_per_second": 4892826.95439903,
+    "generation_bytes_per_second": 299920.63800105173,
+    "upload_bytes_per_second": 4796130.166972937,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.613703635961414,
+    "estimated_full_shards_per_hour": 64.35598711990036,
     "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": 1.3206068420004158,
-      "bytes": 11244076,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.1782944210026471,
       "bytes": 2347277,
@@ -562,6 +556,12 @@
       "bytes": 7678717,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.9852680410003813,
+      "bytes": 917014,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -279,3 +279,4 @@
 {"step": 912900, "loss": 0.5653138756752014, "lr": 3e-05, "grad_norm": 0.3223837000785098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133622263168, "tokens_per_second": 152633.49246152505, "data_wait_s": 1.792699913494289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913000, "loss": 0.5310258269309998, "lr": 3e-05, "grad_norm": 0.3406600564195361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133636845513, "tokens_per_second": 151504.35586913186, "data_wait_s": 1.4197001291904598e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913100, "loss": 0.5355512499809265, "lr": 3e-05, "grad_norm": 0.33200545651987756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133651473372, "tokens_per_second": 152255.45908432166, "data_wait_s": 1.3250999472802505e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 913200, "loss": 0.5601001381874084, "lr": 3e-05, "grad_norm": 0.3221489317124942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133666160476, "tokens_per_second": 153073.64850700658, "data_wait_s": 1.756399797159247e-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:e2c02bf35393ec8662be1df29c5ed5b98f0a88f0aace0357e8317758bf02aaa7
-size 599238
+oid sha256:62155ef73a3493b20d7c031ca6d57032a4a738f7058ac21c58a075786079b328
+size 601356