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": 77536.06780480912,
-      "required_pages_per_second_upper_estimate": 0.0053641250792005515
+      "required_token_positions_per_second": 77370.71325617614,
+      "required_pages_per_second_upper_estimate": 0.005352685467850709
     },
     {
       "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": 9692.00847560114,
-      "required_pages_per_second_upper_estimate": 0.004057841556582448
+      "required_token_positions_per_second": 9671.339157022017,
+      "required_pages_per_second_upper_estimate": 0.0040491877445178225
     },
     {
       "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": 19384.01695120228,
-      "required_pages_per_second_upper_estimate": 0.004037179540032332
+      "required_token_positions_per_second": 19342.678314044035,
+      "required_pages_per_second_upper_estimate": 0.0040285697920854424
     },
     {
       "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": 29076.025426803422,
-      "required_pages_per_second_upper_estimate": 0.0017430633278166803
+      "required_token_positions_per_second": 29014.017471066054,
+      "required_pages_per_second_upper_estimate": 0.0017393460455508913
     },
     {
       "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": 29076.025426803422,
-      "required_pages_per_second_upper_estimate": 0.0022605330850141484
+      "required_token_positions_per_second": 29014.017471066054,
+      "required_pages_per_second_upper_estimate": 0.0022557122391997407
     },
     {
       "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": 29076.025426803422,
-      "required_pages_per_second_upper_estimate": 0.37295603477127565
+      "required_token_positions_per_second": 29014.017471066054,
+      "required_pages_per_second_upper_estimate": 0.3721606632940323
     }
   ],
   "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": 0.1344985748358951,
-    "producer_p95_seconds": 0.16534719299988865,
-    "gpu_update_seconds": 0.9508864982114436,
-    "producer_batches_per_second": 7.4350230195384865,
-    "gpu_batches_per_second": 1.0516502252171376,
-    "producer_headroom": 7.069863003170425,
+    "producer_batch_seconds": 0.3743586325234105,
+    "producer_p95_seconds": 0.1576167049997821,
+    "gpu_update_seconds": 0.9529187065380277,
+    "producer_batches_per_second": 2.6712353158771225,
+    "gpu_batches_per_second": 1.0494074606143682,
+    "producer_headroom": 2.545470102064327,
     "sustainable": true,
-    "buffer_seconds": 60.85673588553239,
+    "buffer_seconds": 60.03387851189574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302532.74159638135,
-    "upload_bytes_per_second": 5266333.658349814,
+    "generation_bytes_per_second": 284169.7022578636,
+    "upload_bytes_per_second": 5240881.068836493,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.91648374479847,
+    "estimated_full_shards_per_hour": 60.97620297243052,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 4727222,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 2.7728241000004346,
+      "bytes": 13625242,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 400700, "loss": 0.6032090187072754, "lr": 0.0003, "grad_norm": 0.1455071352752887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58651151433, "tokens_per_second": 151886.13541794656, "data_wait_s": 1.004999967335607e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 400800, "loss": 0.5677890181541443, "lr": 0.0003, "grad_norm": 0.13917895634959468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58665852294, "tokens_per_second": 153968.55408491456, "data_wait_s": 1.1303000064799562e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 400900, "loss": 0.5627261400222778, "lr": 0.0003, "grad_norm": 0.16006537105349405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58680561683, "tokens_per_second": 152460.9203936658, "data_wait_s": 9.41600046644453e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 401000, "loss": 0.5702412128448486, "lr": 0.0003, "grad_norm": 0.1436486818860728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58695149928, "tokens_per_second": 152054.28080523902, "data_wait_s": 9.996000699175056e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7967e4d89fa608a4b5aee1fff4c09e35593447e7d19b82ecbd37a55934ab32a3
-size 81432
+oid sha256:8d30e20b63b9b62fbaf0579ec43ab6cc9eb158c17dcb6e39dbb0d0559fa28d42
+size 83550