Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75279.04772546975,
-      "required_pages_per_second_upper_estimate": 0.006524997063841572
+      "required_token_positions_per_second": 75816.59753640718,
+      "required_pages_per_second_upper_estimate": 0.006571590519046089
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9409.880965683718,
-      "required_pages_per_second_upper_estimate": 0.003753742207469171
+      "required_token_positions_per_second": 9477.074692050897,
+      "required_pages_per_second_upper_estimate": 0.0037805467895527752
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18819.761931367437,
-      "required_pages_per_second_upper_estimate": 0.004353620706324312
+      "required_token_positions_per_second": 18954.149384101795,
+      "required_pages_per_second_upper_estimate": 0.004384708878376018
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28229.64289705115,
-      "required_pages_per_second_upper_estimate": 0.0017190371429406388
+      "required_token_positions_per_second": 28431.224076152695,
+      "required_pages_per_second_upper_estimate": 0.001731312379133213
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28229.64289705115,
-      "required_pages_per_second_upper_estimate": 0.002173196054868017
+      "required_token_positions_per_second": 28431.224076152695,
+      "required_pages_per_second_upper_estimate": 0.002188714332047672
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28229.64289705115,
-      "required_pages_per_second_upper_estimate": 0.350517686244225
+      "required_token_positions_per_second": 28431.224076152695,
+      "required_pages_per_second_upper_estimate": 0.35302064983985865
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.12144957289063996,
-    "producer_p95_seconds": 0.390962165001838,
-    "gpu_update_seconds": 0.9793960235638719,
-    "producer_batches_per_second": 8.233870043334417,
-    "gpu_batches_per_second": 1.0210374311722783,
-    "producer_headroom": 8.064219578983414,
+    "producer_batch_seconds": 0.2189116486716216,
+    "producer_p95_seconds": 0.39063410100061446,
+    "gpu_update_seconds": 0.9724519748409411,
+    "producer_batches_per_second": 4.568052938562671,
+    "gpu_batches_per_second": 1.0283284171062173,
+    "producer_headroom": 4.442212101283233,
     "sustainable": true,
-    "buffer_seconds": 60.72255346096006,
+    "buffer_seconds": 60.29202244013835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 510752.41340381996,
-    "upload_bytes_per_second": 5583062.491875187,
+    "generation_bytes_per_second": 527936.4549714911,
+    "upload_bytes_per_second": 5716853.5664507225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.59557820878933,
+    "estimated_full_shards_per_hour": 113.28287350519705,
     "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": 0.9720597020022979,
-      "bytes": 1226766,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.4056276539995451,
       "bytes": 7428876,
@@ -562,6 +556,12 @@
       "bytes": 2360873,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.193771635000303,
+      "bytes": 12283347,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 394600, "loss": 0.6006249086931348, "lr": 0.0003, "grad_norm": 0.16905402862110158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57757866352, "tokens_per_second": 146327.3491238891, "data_wait_s": 1.1219999578315765e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394700, "loss": 0.5499810837209225, "lr": 0.0003, "grad_norm": 0.1470346524536101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57772545742, "tokens_per_second": 149039.21257245296, "data_wait_s": 9.467999916523695e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 394800, "loss": 0.593715151771903, "lr": 0.0003, "grad_norm": 0.17916264473158008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57787309955, "tokens_per_second": 151255.48103643116, "data_wait_s": 1.2794000213034451e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 394900, "loss": 0.6093955338001251, "lr": 0.0003, "grad_norm": 0.16123570419690397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57802033460, "tokens_per_second": 149377.65332230096, "data_wait_s": 1.1211996024940163e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50230ae13370b43b22ff75060a277c4c17b74b81374bbc1246eeec9427c78782
-size 848610
+oid sha256:5c238ce268df066ef25dff4ea061a2e8b4c89b83f32c3a7d339aebeec9bcac5b
+size 850728