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": 77498.10062179565,
-      "required_pages_per_second_upper_estimate": 0.005361498421383677
+      "required_token_positions_per_second": 77277.50332981213,
+      "required_pages_per_second_upper_estimate": 0.00534623698886802
     },
     {
       "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": 9687.262577724456,
-      "required_pages_per_second_upper_estimate": 0.004055854548247097
+      "required_token_positions_per_second": 9659.687916226516,
+      "required_pages_per_second_upper_estimate": 0.0040443096133023215
     },
     {
       "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": 19374.52515544891,
-      "required_pages_per_second_upper_estimate": 0.004035202649292392
+      "required_token_positions_per_second": 19319.37583245303,
+      "required_pages_per_second_upper_estimate": 0.0040237164996978015
     },
     {
       "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": 29061.787733173365,
-      "required_pages_per_second_upper_estimate": 0.0017422097998232577
+      "required_token_positions_per_second": 28979.063748679546,
+      "required_pages_per_second_upper_estimate": 0.001737250623265064
     },
     {
       "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": 29061.787733173365,
-      "required_pages_per_second_upper_estimate": 0.0022594261669594074
+      "required_token_positions_per_second": 28979.063748679546,
+      "required_pages_per_second_upper_estimate": 0.0022529947410293663
     },
     {
       "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": 29061.787733173365,
-      "required_pages_per_second_upper_estimate": 0.3727734089246337
+      "required_token_positions_per_second": 28979.063748679546,
+      "required_pages_per_second_upper_estimate": 0.3717123144736413
     }
   ],
   "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": 0.11134604999227804,
-    "producer_p95_seconds": 0.3255206570029259,
-    "gpu_update_seconds": 0.9513523481021245,
-    "producer_batches_per_second": 8.981010103810158,
-    "gpu_batches_per_second": 1.051135262339893,
-    "producer_headroom": 8.5441050505887,
+    "producer_batch_seconds": 0.10599283567216844,
+    "producer_p95_seconds": 0.32410449600138236,
+    "gpu_update_seconds": 0.9540680899761574,
+    "producer_batches_per_second": 9.434599929876011,
+    "gpu_batches_per_second": 1.0481432200766618,
+    "producer_headroom": 9.001250734785996,
     "sustainable": true,
-    "buffer_seconds": 60.88655027853597,
+    "buffer_seconds": 60.10628966849792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 589707.9246721719,
+    "generation_bytes_per_second": 587888.1251244326,
     "upload_bytes_per_second": 6000240.386502057,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.26879646837172,
+    "estimated_full_shards_per_hour": 63.07355315828197,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 515900, "loss": 0.6434698104858398, "lr": 0.0003, "grad_norm": 0.1668728827510193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75510359290, "tokens_per_second": 153016.33690597114, "data_wait_s": 9.945004421751946e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 516000, "loss": 0.5789161920547485, "lr": 0.0003, "grad_norm": 0.14989703641207358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75524952983, "tokens_per_second": 152554.9125770826, "data_wait_s": 1.465899549657479e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516100, "loss": 0.4874710142612457, "lr": 0.0003, "grad_norm": 0.1608227913713671, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75539491589, "tokens_per_second": 151247.0249344127, "data_wait_s": 1.708599302219227e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 516200, "loss": 0.5786738991737366, "lr": 0.0003, "grad_norm": 0.15992800576542088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75553957353, "tokens_per_second": 150329.51372906312, "data_wait_s": 1.6395999409724027e-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 sha256:64e5a1637c89190688440eae1731caed15f7a28dcb3eacf8ca8a4665f2179290
-size 663561
+oid sha256:2fee6d3960f5f92f333c7221f5625e2434a24223f82a140acc084d862b146562
+size 665679