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": 77440.04646116901,
-      "required_pages_per_second_upper_estimate": 0.0053574820998473155
+      "required_token_positions_per_second": 77145.72717431393,
+      "required_pages_per_second_upper_estimate": 0.005337120408667848
     },
     {
       "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": 9680.005807646126,
-      "required_pages_per_second_upper_estimate": 0.004052816290153892
+      "required_token_positions_per_second": 9643.215896789241,
+      "required_pages_per_second_upper_estimate": 0.004037413122738815
     },
     {
       "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": 19360.011615292253,
-      "required_pages_per_second_upper_estimate": 0.004032179861625553
+      "required_token_positions_per_second": 19286.431793578482,
+      "required_pages_per_second_upper_estimate": 0.004016855125192962
     },
     {
       "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": 29040.017422938377,
-      "required_pages_per_second_upper_estimate": 0.0017409047029660093
+      "required_token_positions_per_second": 28929.64769036772,
+      "required_pages_per_second_upper_estimate": 0.0017342882060232222
     },
     {
       "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": 29040.017422938377,
-      "required_pages_per_second_upper_estimate": 0.0022577336210961117
+      "required_token_positions_per_second": 28929.64769036772,
+      "required_pages_per_second_upper_estimate": 0.002249152859853891
     },
     {
       "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": 29040.017422938377,
-      "required_pages_per_second_upper_estimate": 0.3724941627600772
+      "required_token_positions_per_second": 28929.64769036772,
+      "required_pages_per_second_upper_estimate": 0.3710784583364467
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08408266364875772,
-    "producer_p95_seconds": 0.29698451299918815,
-    "gpu_update_seconds": 0.9520655444979575,
-    "producer_batches_per_second": 11.893058052696153,
-    "gpu_batches_per_second": 1.0503478523921577,
-    "producer_headroom": 11.32297079068598,
+    "producer_batch_seconds": 0.13469518042990103,
+    "producer_p95_seconds": 0.3127284850052092,
+    "gpu_update_seconds": 0.955697777446683,
+    "producer_batches_per_second": 7.4241706110667165,
+    "gpu_batches_per_second": 1.0463558915786937,
+    "producer_headroom": 7.095263352381444,
     "sustainable": true,
-    "buffer_seconds": 60.93219484786928,
+    "buffer_seconds": 60.208959979141035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322738.65769707476,
-    "upload_bytes_per_second": 5460237.104724174,
+    "generation_bytes_per_second": 324893.47280671203,
+    "upload_bytes_per_second": 5321755.656743877,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.2522029703539,
+    "estimated_full_shards_per_hour": 69.71457613135358,
     "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": 1.2082140299971798,
-      "bytes": 13368136,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.2696354079998855,
       "bytes": 6070118,
@@ -562,6 +556,12 @@
       "bytes": 6286319,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9432465139980195,
+      "bytes": 858963,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -370,3 +370,4 @@
 {"step": 522000, "loss": 0.6028584837913513, "lr": 0.0003, "grad_norm": 0.17199760238003273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76403168891, "tokens_per_second": 153939.4876438651, "data_wait_s": 1.3580000086221844e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522100, "loss": 0.5927833318710327, "lr": 0.0003, "grad_norm": 0.15929691437666202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76417899080, "tokens_per_second": 154297.78805688574, "data_wait_s": 1.1994001397397369e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522200, "loss": 0.5561525821685791, "lr": 0.0003, "grad_norm": 0.15404037372103327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76432602826, "tokens_per_second": 153506.2498205698, "data_wait_s": 1.4517005183734e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 522300, "loss": 0.5395720601081848, "lr": 0.0003, "grad_norm": 0.1544161777385268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76447122635, "tokens_per_second": 151519.94599900013, "data_wait_s": 1.3866003428120166e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c9be3d803328e1fd1160d6da61c581f5c4bb3910990530c02036fe4957865c8
-size 793542
+oid sha256:69436de78fd038381dcb8ab3b7059d9d136218491b0b602f653ad93e56cc9696
+size 795660