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": 76028.88120803001,
-      "required_pages_per_second_upper_estimate": 0.005259854413280654
+      "required_token_positions_per_second": 76208.1004162286,
+      "required_pages_per_second_upper_estimate": 0.005272253213949688
     },
     {
       "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": 9503.610151003752,
-      "required_pages_per_second_upper_estimate": 0.003978963112277912
+      "required_token_positions_per_second": 9526.012552028575,
+      "required_pages_per_second_upper_estimate": 0.003988342529771675
     },
     {
       "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": 19007.220302007503,
-      "required_pages_per_second_upper_estimate": 0.00395870273480092
+      "required_token_positions_per_second": 19052.02510405715,
+      "required_pages_per_second_upper_estimate": 0.003968034393485773
     },
     {
       "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": 28510.830453011255,
-      "required_pages_per_second_upper_estimate": 0.0017091807521405955
+      "required_token_positions_per_second": 28578.037656085722,
+      "required_pages_per_second_upper_estimate": 0.0017132097213454524
     },
     {
       "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": 28510.830453011255,
-      "required_pages_per_second_upper_estimate": 0.002216591661831765
+      "required_token_positions_per_second": 28578.037656085722,
+      "required_pages_per_second_upper_estimate": 0.0022218167262575603
     },
     {
       "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": 28510.830453011255,
-      "required_pages_per_second_upper_estimate": 0.3657063205065514
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28578.037656085722,
+      "required_pages_per_second_upper_estimate": 0.3665683823461182
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.35333880423428354,
-    "producer_p95_seconds": 0.29952018999756547,
-    "gpu_update_seconds": 0.9697367477796451,
-    "producer_batches_per_second": 2.83014485818247,
-    "gpu_batches_per_second": 1.0312076986766223,
-    "producer_headroom": 2.7444954705191535,
+    "producer_batch_seconds": 0.10301916209351702,
+    "producer_p95_seconds": 0.30279377099941485,
+    "gpu_update_seconds": 0.9674562100002108,
+    "producer_batches_per_second": 9.706931988946257,
+    "gpu_batches_per_second": 1.0336385147600449,
+    "producer_headroom": 9.391031632755753,
     "sustainable": true,
-    "buffer_seconds": 60.123678362338,
+    "buffer_seconds": 60.94974123001328,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450064.38035138237,
-    "upload_bytes_per_second": 5081976.493351355,
+    "generation_bytes_per_second": 449776.84152971616,
+    "upload_bytes_per_second": 5121892.579817094,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.57333906084159,
+    "estimated_full_shards_per_hour": 96.51163992327322,
     "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": 1.2359630550017755,
-      "bytes": 7977289,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.4249079500004882,
       "bytes": 10392834,
@@ -562,6 +556,12 @@
       "bytes": 2380554,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3700068609978189,
+      "bytes": 12013278,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -342,3 +342,4 @@
 {"step": 754200, "loss": 0.5593996644020081, "lr": 5e-05, "grad_norm": 0.23872154556980332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110394023411, "tokens_per_second": 150613.9667814889, "data_wait_s": 1.1531003110576421e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 754300, "loss": 0.5498316884040833, "lr": 5e-05, "grad_norm": 0.2421564901006812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110408529030, "tokens_per_second": 149676.33542256468, "data_wait_s": 1.0902003850787878e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 754400, "loss": 0.6513687372207642, "lr": 5e-05, "grad_norm": 0.3011589707814446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110423082674, "tokens_per_second": 150632.93927386892, "data_wait_s": 1.210599293699488e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 754500, "loss": 0.5053891539573669, "lr": 5e-05, "grad_norm": 0.24702827742008185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110437709006, "tokens_per_second": 151278.6627559164, "data_wait_s": 9.224000677932054e-06, "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:fe8df90bf9f7194025c0667f8a2f51732baed0e8ff0513d6069e86713f5eda48
-size 733455
+oid sha256:36ceed55bd7ade0733db4db85c78c8e4331bb955af3ff7611f561575928c1b27
+size 735573