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": 75556.08261414293,
-      "required_pages_per_second_upper_estimate": 0.005227145109511664
+      "required_token_positions_per_second": 76338.38911233666,
+      "required_pages_per_second_upper_estimate": 0.005281266888257869
     },
     {
       "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": 9444.510326767866,
-      "required_pages_per_second_upper_estimate": 0.003954219250015017
+      "required_token_positions_per_second": 9542.298639042083,
+      "required_pages_per_second_upper_estimate": 0.003995161174312061
     },
     {
       "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": 18889.020653535732,
-      "required_pages_per_second_upper_estimate": 0.003934084865158598
+      "required_token_positions_per_second": 19084.597278084166,
+      "required_pages_per_second_upper_estimate": 0.003974818318349608
     },
     {
       "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": 28333.530980303596,
-      "required_pages_per_second_upper_estimate": 0.0016985519194723908
+      "required_token_positions_per_second": 28626.89591712625,
+      "required_pages_per_second_upper_estimate": 0.0017161386994925872
     },
     {
       "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": 28333.530980303596,
-      "required_pages_per_second_upper_estimate": 0.0022028074076867075
+      "required_token_positions_per_second": 28626.89591712625,
+      "required_pages_per_second_upper_estimate": 0.002225615241148683
     },
     {
       "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": 28333.530980303596,
-      "required_pages_per_second_upper_estimate": 0.3634321132399994
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28626.89591712625,
+      "required_pages_per_second_upper_estimate": 0.36719508365883263
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11943939625777489,
-    "producer_p95_seconds": 0.24581599300108792,
-    "gpu_update_seconds": 0.9758049577096426,
-    "producer_batches_per_second": 8.372446875416161,
-    "gpu_batches_per_second": 1.0247949573315827,
-    "producer_headroom": 8.169875169191696,
+    "producer_batch_seconds": 0.06157200333601054,
+    "producer_p95_seconds": 0.21132788399881974,
+    "gpu_update_seconds": 0.9658050275531055,
+    "producer_batches_per_second": 16.241147694071334,
+    "gpu_batches_per_second": 1.035405668298837,
+    "producer_headroom": 15.68578209616662,
     "sustainable": true,
-    "buffer_seconds": 60.49990737799784,
+    "buffer_seconds": 60.84571673584565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449038.8544759854,
+    "generation_bytes_per_second": 471013.7399382933,
     "upload_bytes_per_second": 3927811.8192784693,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.35328508100196,
+    "estimated_full_shards_per_hour": 101.06858395206069,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 814500, "loss": 0.5320994257926941, "lr": 3e-05, "grad_norm": 0.2902150563963192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119219658989, "tokens_per_second": 152389.13778780532, "data_wait_s": 1.1037000149372034e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 814600, "loss": 0.5440930128097534, "lr": 3e-05, "grad_norm": 0.2911424202091485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119234400162, "tokens_per_second": 152597.91513367646, "data_wait_s": 1.3880000551580451e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 814700, "loss": 0.5845077633857727, "lr": 3e-05, "grad_norm": 0.28333416091779967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119249129344, "tokens_per_second": 152526.47696780687, "data_wait_s": 1.3072000001557171e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 814800, "loss": 0.5097643733024597, "lr": 3e-05, "grad_norm": 0.2702926412809111, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119263684066, "tokens_per_second": 149738.98734738777, "data_wait_s": 7.728000127826817e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f523f622725e365766db7e9bd933cfb2fba0e037559f43363b6dfc1b6f2b5c6
-size 313077
+oid sha256:f402a9b208a9a88ae9b610a6f237c43b9a6bcf47619d430f4173de81a1a19eeb
+size 315195