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": 77601.31467707505,
-      "required_pages_per_second_upper_estimate": 0.005368639009217509
+      "required_token_positions_per_second": 77380.12612121923,
+      "required_pages_per_second_upper_estimate": 0.005353336671695246
     },
     {
       "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": 9700.16433463438,
-      "required_pages_per_second_upper_estimate": 0.004061256244445963
+      "required_token_positions_per_second": 9672.515765152404,
+      "required_pages_per_second_upper_estimate": 0.004049680365771644
     },
     {
       "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": 19400.32866926876,
-      "required_pages_per_second_upper_estimate": 0.0040405768407366476
+      "required_token_positions_per_second": 19345.03153030481,
+      "required_pages_per_second_upper_estimate": 0.0040290599049740755
     },
     {
       "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": 29100.493003903142,
-      "required_pages_per_second_upper_estimate": 0.0017445301217039815
+      "required_token_positions_per_second": 29017.54729545721,
+      "required_pages_per_second_upper_estimate": 0.0017395576531334114
     },
     {
       "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": 29100.493003903142,
-      "required_pages_per_second_upper_estimate": 0.002262435331512154
+      "required_token_positions_per_second": 29017.54729545721,
+      "required_pages_per_second_upper_estimate": 0.0022559866675200967
     },
     {
       "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": 29100.493003903142,
-      "required_pages_per_second_upper_estimate": 0.3732698785790734
+      "required_token_positions_per_second": 29017.54729545721,
+      "required_pages_per_second_upper_estimate": 0.37220594009129193
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4448048943280867,
-    "producer_p95_seconds": 0.29354744000011124,
-    "gpu_update_seconds": 0.9500869966804918,
-    "producer_batches_per_second": 2.2481767011817166,
-    "gpu_batches_per_second": 1.0525351925601543,
-    "producer_headroom": 2.1359634500327926,
+    "producer_batch_seconds": 0.0772295542578263,
+    "producer_p95_seconds": 0.2756188950006617,
+    "gpu_update_seconds": 0.9528027892394745,
+    "producer_batches_per_second": 12.9484108721068,
+    "gpu_batches_per_second": 1.049535130767405,
+    "producer_headroom": 12.337281995162096,
     "sustainable": true,
-    "buffer_seconds": 60.805567787551475,
+    "buffer_seconds": 60.02657572208689,
     "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": 283664.162194505,
+    "generation_bytes_per_second": 311678.1857829196,
     "upload_bytes_per_second": 3963764.523420762,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.867725843204134,
+    "estimated_full_shards_per_hour": 66.87888317218486,
     "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-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -304,3 +304,4 @@
 {"step": 555400, "loss": 0.6096173524856567, "lr": 0.0003, "grad_norm": 0.15936637297269335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81292797234, "tokens_per_second": 153260.81056322544, "data_wait_s": 1.4658999134553596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 555500, "loss": 0.6384610533714294, "lr": 0.0003, "grad_norm": 0.16261030728629372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81307454820, "tokens_per_second": 153254.33135518924, "data_wait_s": 1.7139998817583546e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 555600, "loss": 0.5761717557907104, "lr": 0.0003, "grad_norm": 0.16174715844495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81322033287, "tokens_per_second": 151586.76933585954, "data_wait_s": 1.5345998690463603e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 555700, "loss": 0.6015921831130981, "lr": 0.0003, "grad_norm": 0.20086125779030492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81336580275, "tokens_per_second": 151461.28770380846, "data_wait_s": 1.2984997738385573e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1908fdfe20edd52a7a52abbbad9f5332837bd59b87bf7ea0a4deb0262901e4d
-size 652971
+oid sha256:17d71d631a03ac3046ab872f63d53722730e6000d99a6b12a67efd2958b201b3
+size 655089