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": 76392.81001822512,
-      "required_pages_per_second_upper_estimate": 0.005285031852801132
+      "required_token_positions_per_second": 74392.00616220923,
+      "required_pages_per_second_upper_estimate": 0.005146611599537394
     },
     {
       "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": 9549.10125227814,
-      "required_pages_per_second_upper_estimate": 0.00399800928641928
+      "required_token_positions_per_second": 9299.000770276154,
+      "required_pages_per_second_upper_estimate": 0.003893297437297005
     },
     {
       "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": 19098.20250455628,
-      "required_pages_per_second_upper_estimate": 0.003977651928229799
+      "required_token_positions_per_second": 18598.001540552308,
+      "required_pages_per_second_upper_estimate": 0.0038734732586142616
     },
     {
       "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": 28647.30375683442,
-      "required_pages_per_second_upper_estimate": 0.001717362118322125
+      "required_token_positions_per_second": 27897.002310828462,
+      "required_pages_per_second_upper_estimate": 0.001672382692278041
     },
     {
       "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": 28647.30375683442,
-      "required_pages_per_second_upper_estimate": 0.0022272018609213926
+      "required_token_positions_per_second": 27897.002310828462,
+      "required_pages_per_second_upper_estimate": 0.0021688692237217116
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28647.30375683442,
-      "required_pages_per_second_upper_estimate": 0.36745685351437796
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 27897.002310828462,
+      "required_pages_per_second_upper_estimate": 0.35783279217594005
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0945171626954675,
-    "producer_p95_seconds": 0.2710071359979338,
-    "gpu_update_seconds": 0.9651170048910445,
-    "producer_batches_per_second": 10.580089070404927,
-    "gpu_batches_per_second": 1.036143799075319,
-    "producer_headroom": 10.211023875109678,
+    "producer_batch_seconds": 0.6152925128124593,
+    "producer_p95_seconds": 0.272032927001419,
+    "gpu_update_seconds": 0.9910742269705513,
+    "producer_batches_per_second": 1.6252432447602352,
+    "gpu_batches_per_second": 1.0090061599692008,
+    "producer_headroom": 1.6107366924398605,
     "sustainable": true,
-    "buffer_seconds": 60.802371308135804,
+    "buffer_seconds": 60.455527845203626,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314476.422476751,
-    "upload_bytes_per_second": 4907672.811227513,
+    "generation_bytes_per_second": 276910.0639599688,
+    "upload_bytes_per_second": 5011688.749704315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.47931962706467,
+    "estimated_full_shards_per_hour": 59.418453589432694,
     "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": 1.1885178210013692,
-      "bytes": 1081987,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.9165967070002807,
       "bytes": 10316454,
@@ -562,6 +556,12 @@
       "bytes": 1306429,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.5331175120008993,
+      "bytes": 11738494,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -184,3 +184,4 @@
 {"step": 738400, "loss": 0.5231951475143433, "lr": 5e-05, "grad_norm": 0.24424534864142355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108081583373, "tokens_per_second": 148619.50765769536, "data_wait_s": 1.7316000594291836e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738500, "loss": 0.5642023086547852, "lr": 5e-05, "grad_norm": 0.23152752564406837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108096143363, "tokens_per_second": 149230.83097907345, "data_wait_s": 1.7384001694153994e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738600, "loss": 0.5704699158668518, "lr": 5e-05, "grad_norm": 0.26998518373392777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108110739457, "tokens_per_second": 130833.35308083435, "data_wait_s": 1.2314998457441106e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 738700, "loss": 0.5004143714904785, "lr": 5e-05, "grad_norm": 0.22349828997716606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108125463221, "tokens_per_second": 151822.54949203963, "data_wait_s": 1.803800114430487e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc13c31760847a9df0f746295b72171ee4ba1e8b362d13fe32c9db860e08f586
-size 396462
+oid sha256:16cf6cc43cf11dcd0429ef079cf5aac816a91299b931bb2730078ee361753413
+size 398580