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": 77384.20293710922,
-      "required_pages_per_second_upper_estimate": 0.005353618715278031
+      "required_token_positions_per_second": 76141.60687187794,
+      "required_pages_per_second_upper_estimate": 0.00526765303626523
     },
     {
       "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": 9673.025367138653,
-      "required_pages_per_second_upper_estimate": 0.004049893725481587
+      "required_token_positions_per_second": 9517.700858984743,
+      "required_pages_per_second_upper_estimate": 0.003984862597462111
     },
     {
       "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": 19346.050734277305,
-      "required_pages_per_second_upper_estimate": 0.00402927217828333
+      "required_token_positions_per_second": 19035.401717969486,
+      "required_pages_per_second_upper_estimate": 0.003964572180551885
     },
     {
       "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": 29019.076101415958,
-      "required_pages_per_second_upper_estimate": 0.0017396493027162867
+      "required_token_positions_per_second": 28553.10257695423,
+      "required_pages_per_second_upper_estimate": 0.0017117149014251808
     },
     {
       "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": 29019.076101415958,
-      "required_pages_per_second_upper_estimate": 0.0022561055254588826
+      "required_token_positions_per_second": 28553.10257695423,
+      "required_pages_per_second_upper_estimate": 0.0022198781335329085
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 29019.076101415958,
-      "required_pages_per_second_upper_estimate": 0.3722255499726268
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28553.10257695423,
+      "required_pages_per_second_upper_estimate": 0.36624854192422146
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11542594855472998,
-    "producer_p95_seconds": 0.2167059330004122,
-    "gpu_update_seconds": 0.952752592928034,
-    "producer_batches_per_second": 8.663563198060645,
-    "gpu_batches_per_second": 1.0495904261218156,
-    "producer_headroom": 8.254232300948171,
+    "producer_batch_seconds": 0.22824270127343027,
+    "producer_p95_seconds": 0.22941520899985335,
+    "gpu_update_seconds": 0.9683010778070488,
+    "producer_batches_per_second": 4.381301108078017,
+    "gpu_batches_per_second": 1.0327366383446985,
+    "producer_headroom": 4.242418585149162,
     "sustainable": true,
-    "buffer_seconds": 60.02341335446614,
+    "buffer_seconds": 60.034666824037025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 518993.0865720847,
+    "generation_bytes_per_second": 491050.5838319743,
     "upload_bytes_per_second": 5112430.952778343,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.36383483764558,
+    "estimated_full_shards_per_hour": 105.36802421779082,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 455200, "loss": 0.5711075067520142, "lr": 0.0003, "grad_norm": 0.1762686802571171, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66627342094, "tokens_per_second": 152355.7289531545, "data_wait_s": 1.1676000212901272e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455300, "loss": 0.58512943983078, "lr": 0.0003, "grad_norm": 0.18817986255199826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66642029194, "tokens_per_second": 153155.42990572797, "data_wait_s": 1.2401998901623301e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455400, "loss": 0.6065024137496948, "lr": 0.0003, "grad_norm": 0.1634300136332109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66656557049, "tokens_per_second": 152039.80700311853, "data_wait_s": 1.4707000445923768e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 455500, "loss": 0.6034132242202759, "lr": 0.0003, "grad_norm": 0.1672609640580655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66671161434, "tokens_per_second": 150856.0205691735, "data_wait_s": 1.2145999789936468e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c7c857df29228f188086d19f3846c6c3736d3b163d7cb9e47bc843a7a372a07
-size 324450
+oid sha256:5870d2a62b460d8dd8597a581d166b06e392646d5819a95a9b437e3aab62ea7b
+size 326568