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": 76396.56749591012,
-      "required_pages_per_second_upper_estimate": 0.005285291803825929
+      "required_token_positions_per_second": 76070.05578647502,
+      "required_pages_per_second_upper_estimate": 0.005262702966155664
     },
     {
       "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": 9549.570936988765,
-      "required_pages_per_second_upper_estimate": 0.003998205933599487
+      "required_token_positions_per_second": 9508.756973309377,
+      "required_pages_per_second_upper_estimate": 0.0039811179792994065
     },
     {
       "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": 19099.14187397753,
-      "required_pages_per_second_upper_estimate": 0.003977847574107408
+      "required_token_positions_per_second": 19017.513946618754,
+      "required_pages_per_second_upper_estimate": 0.00396084662951178
     },
     {
       "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": 28648.71281096629,
-      "required_pages_per_second_upper_estimate": 0.0017174465889658295
+      "required_token_positions_per_second": 28526.270919928127,
+      "required_pages_per_second_upper_estimate": 0.001710106384556034
     },
     {
       "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": 28648.71281096629,
-      "required_pages_per_second_upper_estimate": 0.0022273114086823714
+      "required_token_positions_per_second": 28526.270919928127,
+      "required_pages_per_second_upper_estimate": 0.0022177920902191157
     },
     {
       "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": 28648.71281096629,
-      "required_pages_per_second_upper_estimate": 0.36747492734785714
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28526.270919928127,
+      "required_pages_per_second_upper_estimate": 0.36590437423747935
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10753726667977048,
-    "producer_p95_seconds": 0.25298658500105375,
-    "gpu_update_seconds": 0.9650695367163848,
-    "producer_batches_per_second": 9.29910189160607,
-    "gpu_batches_per_second": 1.0361947631281212,
-    "producer_headroom": 8.974279954410727,
+    "producer_batch_seconds": 0.32857051441425256,
+    "producer_p95_seconds": 0.26346257499972126,
+    "gpu_update_seconds": 0.9692118565937555,
+    "producer_batches_per_second": 3.0434867285115788,
+    "gpu_batches_per_second": 1.0317661646386043,
+    "producer_headroom": 2.9497834226591624,
     "sustainable": true,
-    "buffer_seconds": 60.799380813132245,
+    "buffer_seconds": 60.09113510881284,
     "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": 341495.7370970572,
+    "generation_bytes_per_second": 324333.6572155814,
     "upload_bytes_per_second": 5387116.212514579,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.27703556712902,
+    "estimated_full_shards_per_hour": 69.59445273733694,
     "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": [

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

@@ -254,3 +254,4 @@
 {"step": 745400, "loss": 0.5447136759757996, "lr": 5e-05, "grad_norm": 0.22656630798232283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109105958134, "tokens_per_second": 147603.6770643994, "data_wait_s": 1.3957996998215094e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745500, "loss": 0.5598964095115662, "lr": 5e-05, "grad_norm": 0.2301542085667018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109120644051, "tokens_per_second": 151447.52547444103, "data_wait_s": 1.556699862703681e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745600, "loss": 0.5423048138618469, "lr": 5e-05, "grad_norm": 0.23486748478399427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109135360963, "tokens_per_second": 151800.3414922986, "data_wait_s": 1.3477998436428607e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 745700, "loss": 0.5213366746902466, "lr": 5e-05, "grad_norm": 0.22292530404676922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109150134328, "tokens_per_second": 151497.57275360494, "data_wait_s": 1.5575998986605555e-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 sha256:54dec80053c59b2ea78cfb1e98eda8f000434fd803979c6db2315ecf318b0c0e
-size 546288
+oid sha256:06dbbcd1c34c1b1fe89fae81b255a9a8142dbe207ecadfde689d42bb34bd1607
+size 548406