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": 77531.22937080052,
-      "required_pages_per_second_upper_estimate": 0.005363790345109122
+      "required_token_positions_per_second": 76296.39665152293,
+      "required_pages_per_second_upper_estimate": 0.0052783617523828414
     },
     {
       "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": 9691.403671350065,
-      "required_pages_per_second_upper_estimate": 0.004057588337672271
+      "required_token_positions_per_second": 9537.049581440366,
+      "required_pages_per_second_upper_estimate": 0.003992963503507009
     },
     {
       "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": 19382.80734270013,
-      "required_pages_per_second_upper_estimate": 0.004036927610480856
+      "required_token_positions_per_second": 19074.09916288073,
+      "required_pages_per_second_upper_estimate": 0.003972631837806648
     },
     {
       "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": 29074.211014050194,
-      "required_pages_per_second_upper_estimate": 0.0017429545565425721
+      "required_token_positions_per_second": 28611.148744321094,
+      "required_pages_per_second_upper_estimate": 0.0017151946805274586
     },
     {
       "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": 29074.211014050194,
-      "required_pages_per_second_upper_estimate": 0.002260392022403343
+      "required_token_positions_per_second": 28611.148744321094,
+      "required_pages_per_second_upper_estimate": 0.002224390967727575
     },
     {
       "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": 29074.211014050194,
-      "required_pages_per_second_upper_estimate": 0.37293276143264187
+      "required_token_positions_per_second": 28611.148744321094,
+      "required_pages_per_second_upper_estimate": 0.3669930958340849
     }
   ],
   "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": 0.08790938617238453,
-    "producer_p95_seconds": 0.2963143509987276,
-    "gpu_update_seconds": 0.9509458394808729,
-    "producer_batches_per_second": 11.375349590532526,
-    "gpu_batches_per_second": 1.0515845997558664,
-    "producer_headroom": 10.817341365757356,
+    "producer_batch_seconds": 0.9118546186877552,
+    "producer_p95_seconds": 0.4556006439961493,
+    "gpu_update_seconds": 0.9663365930208494,
+    "producer_batches_per_second": 1.0966660468738914,
+    "gpu_batches_per_second": 1.0348361090972618,
+    "producer_headroom": 1.0597485314177593,
     "sustainable": true,
-    "buffer_seconds": 60.86053372677586,
+    "buffer_seconds": 60.87920536031351,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300505.0037125466,
-    "upload_bytes_per_second": 5330260.818586267,
+    "generation_bytes_per_second": 254666.2865500591,
+    "upload_bytes_per_second": 5330419.314515897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.48137839824962,
+    "estimated_full_shards_per_hour": 54.6454567659028,
     "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": 1.050476451000577,
-      "bytes": 6070361,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0647863270023663,
       "bytes": 1121901,
@@ -562,6 +556,12 @@
       "bytes": 6147230,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.4379745080004795,
+      "bytes": 8148657,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -380,3 +380,4 @@
 {"step": 523000, "loss": 0.5470998883247375, "lr": 0.0003, "grad_norm": 0.1361541966712855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76549333904, "tokens_per_second": 150580.81061198952, "data_wait_s": 1.6494996089022607e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523100, "loss": 0.5512901544570923, "lr": 0.0003, "grad_norm": 0.18323662469231008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76563821788, "tokens_per_second": 151782.80051866503, "data_wait_s": 9.128001693170518e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 523200, "loss": 0.5553374886512756, "lr": 0.0003, "grad_norm": 0.17085884588890812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76578318938, "tokens_per_second": 141690.74500809534, "data_wait_s": 8.859002264216542e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 523300, "loss": 0.5911731123924255, "lr": 0.0003, "grad_norm": 0.1815925787559326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76593048711, "tokens_per_second": 154341.8191260696, "data_wait_s": 1.0716001270338893e-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 sha256:1eb6568589aad7cc25422aa7d9e6424843095cc5b68ce71ca8837998ae60c510
-size 814722
+oid sha256:0a5abeedb667fd556fb5e4c99b226b5222a96c676149c1067f2e1881a1295a66
+size 816840