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": 76028.66108249525,
-      "required_pages_per_second_upper_estimate": 0.005259839184485398
+      "required_token_positions_per_second": 76439.30024512576,
+      "required_pages_per_second_upper_estimate": 0.005288248154570306
     },
     {
       "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": 9503.582635311906,
-      "required_pages_per_second_upper_estimate": 0.00397895159203233
+      "required_token_positions_per_second": 9554.91253064072,
+      "required_pages_per_second_upper_estimate": 0.004000442347316401
     },
     {
       "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": 19007.165270623813,
-      "required_pages_per_second_upper_estimate": 0.003958691273214973
+      "required_token_positions_per_second": 19109.82506128144,
+      "required_pages_per_second_upper_estimate": 0.003980072600288218
     },
     {
       "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": 28510.747905935717,
-      "required_pages_per_second_upper_estimate": 0.001709175803569459
+      "required_token_positions_per_second": 28664.73759192216,
+      "required_pages_per_second_upper_estimate": 0.0017184072501157102
     },
     {
       "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": 28510.747905935717,
-      "required_pages_per_second_upper_estimate": 0.0022165852441597285
+      "required_token_positions_per_second": 28664.73759192216,
+      "required_pages_per_second_upper_estimate": 0.002228557264916128
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28510.747905935717,
-      "required_pages_per_second_upper_estimate": 0.3657052616812986
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28664.73759192216,
+      "required_pages_per_second_upper_estimate": 0.3676804760318898
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2977096596953288,
-    "producer_p95_seconds": 0.30091678100143326,
-    "gpu_update_seconds": 0.969739555455292,
-    "producer_batches_per_second": 3.3589773372600122,
-    "gpu_batches_per_second": 1.0312047130329758,
-    "producer_headroom": 3.257333189818925,
+    "producer_batch_seconds": 0.0955539964376726,
+    "producer_p95_seconds": 0.28068270299991127,
+    "gpu_update_seconds": 0.964530022692108,
+    "producer_batches_per_second": 10.465287034355221,
+    "gpu_batches_per_second": 1.0367743631337587,
+    "producer_headroom": 10.094083540726064,
     "sustainable": true,
-    "buffer_seconds": 60.123852438228106,
+    "buffer_seconds": 60.7653914296028,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482057.50964959944,
-    "upload_bytes_per_second": 5157562.162925388,
+    "generation_bytes_per_second": 476203.0585713808,
+    "upload_bytes_per_second": 5128994.078647128,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.43831984630573,
+    "estimated_full_shards_per_hour": 102.18209092956607,
     "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": 1.1971988570003305,
-      "bytes": 10569382,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8849704640015261,
       "bytes": 2587633,
@@ -562,6 +556,12 @@
       "bytes": 6106429,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9636820249979792,
+      "bytes": 6954429,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -260,3 +260,4 @@
 {"step": 466200, "loss": 0.5670055747032166, "lr": 0.0003, "grad_norm": 0.16899504149586975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68237771772, "tokens_per_second": 152453.98517826578, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466300, "loss": 0.5890834331512451, "lr": 0.0003, "grad_norm": 0.15059000017827742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68252450588, "tokens_per_second": 151839.93936945655, "data_wait_s": 1.9794002582784742e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466400, "loss": 0.5581997632980347, "lr": 0.0003, "grad_norm": 0.15003424692333678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68267135738, "tokens_per_second": 152192.67220994682, "data_wait_s": 1.2128999514970928e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 466500, "loss": 0.5825616717338562, "lr": 0.0003, "grad_norm": 0.16656615871077907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68281815191, "tokens_per_second": 152342.5657942288, "data_wait_s": 2.007999864872545e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cab290ff15b34eddd227b1affcaf32e5f61f1c9c90ef958a5daa6749bc09925
-size 558996
+oid sha256:6a7dc83b90f3eb384216bb95ab64e4f122f349ec3e6cb2b824a234fc2e19cd9a
+size 561114