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": 77626.84721658517,
-      "required_pages_per_second_upper_estimate": 0.005370405409544474
+      "required_token_positions_per_second": 76826.23510929271,
+      "required_pages_per_second_upper_estimate": 0.005315017206285955
     },
     {
       "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": 9703.355902073146,
-      "required_pages_per_second_upper_estimate": 0.004062592487085066
+      "required_token_positions_per_second": 9603.27938866159,
+      "required_pages_per_second_upper_estimate": 0.004020692540754891
     },
     {
       "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": 19406.711804146293,
-      "required_pages_per_second_upper_estimate": 0.00404190627939705
+      "required_token_positions_per_second": 19206.55877732318,
+      "required_pages_per_second_upper_estimate": 0.004000219682300069
     },
     {
       "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": 29110.06770621944,
-      "required_pages_per_second_upper_estimate": 0.0017451041104881197
+      "required_token_positions_per_second": 28809.83816598477,
+      "required_pages_per_second_upper_estimate": 0.0017271058079750155
     },
     {
       "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": 29110.06770621944,
-      "required_pages_per_second_upper_estimate": 0.0022631797225025813
+      "required_token_positions_per_second": 28809.83816598477,
+      "required_pages_per_second_upper_estimate": 0.002239838196319521
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 29110.06770621944,
-      "required_pages_per_second_upper_estimate": 0.3733926925798725
+      "required_token_positions_per_second": 28809.83816598477,
+      "required_pages_per_second_upper_estimate": 0.36954167039910685
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1076554447347462,
-    "producer_p95_seconds": 0.32363712300139014,
-    "gpu_update_seconds": 0.9497744999779899,
-    "producer_batches_per_second": 9.288893863788445,
-    "gpu_batches_per_second": 1.052881499790923,
-    "producer_headroom": 8.822354524828288,
+    "producer_batch_seconds": 0.1271348553984808,
+    "producer_p95_seconds": 0.3210992399981478,
+    "gpu_update_seconds": 0.9596721731204818,
+    "producer_batches_per_second": 7.865663565398208,
+    "gpu_batches_per_second": 1.0420225031099815,
+    "producer_headroom": 7.548458446840295,
     "sustainable": true,
-    "buffer_seconds": 60.785567998591354,
+    "buffer_seconds": 60.459346906590355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333615.85955381073,
-    "upload_bytes_per_second": 5401079.74139039,
+    "generation_bytes_per_second": 345923.98800616147,
+    "upload_bytes_per_second": 5405264.117713781,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.5861972805094,
+    "estimated_full_shards_per_hour": 74.22723512781747,
     "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": 1.2024595770017186,
-      "bytes": 7444430,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.016105377002532,
       "bytes": 1143206,
@@ -562,6 +556,12 @@
       "bytes": 6053776,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3147696169980918,
+      "bytes": 8376347,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 719700, "loss": 0.568727433681488, "lr": 8e-05, "grad_norm": 0.2325424623813383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105344170815, "tokens_per_second": 152920.7540962515, "data_wait_s": 1.1985001037828624e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719800, "loss": 0.5495955348014832, "lr": 8e-05, "grad_norm": 0.21961186793742574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105358880287, "tokens_per_second": 153857.88243120036, "data_wait_s": 1.657800021348521e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719900, "loss": 0.5301546454429626, "lr": 8e-05, "grad_norm": 0.2325197336728661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105373378438, "tokens_per_second": 150446.81822326215, "data_wait_s": 1.312600215896964e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720000, "loss": 0.5157252550125122, "lr": 8e-05, "grad_norm": 0.2200720133218461, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105387890023, "tokens_per_second": 151146.37492137, "data_wait_s": 2.067800232907757e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1113b08792728c5fd091b32c915fff90824e6d0d74227034d96e6c9d21a6d9e7
-size 744045
+oid sha256:38e74344c6bc1d94cfedd0ec25e319aea91cec7e2541f5127251ecc37df5c6b8
+size 746163