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": 76570.57702253615,
-      "required_pages_per_second_upper_estimate": 0.0052973301866356464
+      "required_token_positions_per_second": 76816.62244146192,
+      "required_pages_per_second_upper_estimate": 0.005314352179620958
     },
     {
       "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": 9571.322127817019,
-      "required_pages_per_second_upper_estimate": 0.004007312702982761
+      "required_token_positions_per_second": 9602.07780518274,
+      "required_pages_per_second_upper_estimate": 0.004020189462844213
     },
     {
       "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": 19142.644255634037,
-      "required_pages_per_second_upper_estimate": 0.003986907972971506
+      "required_token_positions_per_second": 19204.15561036548,
+      "required_pages_per_second_upper_estimate": 0.003999719165998556
     },
     {
       "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": 28713.966383451054,
-      "required_pages_per_second_upper_estimate": 0.001721358440999855
+      "required_token_positions_per_second": 28806.23341554822,
+      "required_pages_per_second_upper_estimate": 0.0017268897086904797
     },
     {
       "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": 28713.966383451054,
-      "required_pages_per_second_upper_estimate": 0.0022323845869229295
+      "required_token_positions_per_second": 28806.23341554822,
+      "required_pages_per_second_upper_estimate": 0.0022395579428287006
     },
     {
       "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": 28713.966383451054,
-      "required_pages_per_second_upper_estimate": 0.3683119301118643
+      "required_token_positions_per_second": 28806.23341554822,
+      "required_pages_per_second_upper_estimate": 0.36949543253098627
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3792297348359739,
-    "producer_p95_seconds": 0.3585142510055448,
-    "gpu_update_seconds": 0.9628763797653044,
-    "producer_batches_per_second": 2.63692402820819,
-    "gpu_batches_per_second": 1.038554918382923,
-    "producer_headroom": 2.539031861997245,
+    "producer_batch_seconds": 0.15836826548422778,
+    "producer_p95_seconds": 0.3722331210010452,
+    "gpu_update_seconds": 0.9597922644436027,
+    "producer_batches_per_second": 6.3143963656001025,
+    "gpu_batches_per_second": 1.0418921229581966,
+    "producer_headroom": 6.060508786333777,
     "sustainable": true,
-    "buffer_seconds": 60.661211925214175,
+    "buffer_seconds": 60.46691265994697,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326005.1267398465,
-    "upload_bytes_per_second": 4664858.340085298,
+    "generation_bytes_per_second": 339222.93852560705,
+    "upload_bytes_per_second": 4673546.830929934,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.9531111874251,
+    "estimated_full_shards_per_hour": 72.78934590173873,
     "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": 1.2863249110014294,
-      "bytes": 4967385,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.6688778630013985,
       "bytes": 7600081,
@@ -562,6 +556,12 @@
       "bytes": 6084966,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.6271315569974831,
+      "bytes": 7351886,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -377,3 +377,4 @@
 {"step": 642100, "loss": 0.5516042709350586, "lr": 0.0001, "grad_norm": 0.188053310168775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93986618751, "tokens_per_second": 153111.63567327755, "data_wait_s": 1.1608004570007324e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 642200, "loss": 0.5519996881484985, "lr": 0.0001, "grad_norm": 0.16950966957082045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94001263868, "tokens_per_second": 151976.1155339248, "data_wait_s": 9.807998139876872e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 642300, "loss": 0.5449856519699097, "lr": 0.0001, "grad_norm": 0.2050308693192249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94015954703, "tokens_per_second": 151240.73735878651, "data_wait_s": 1.685600000200793e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 642400, "loss": 0.554408073425293, "lr": 0.0001, "grad_norm": 0.16020420064794735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94030573306, "tokens_per_second": 151646.24168440726, "data_wait_s": 9.57100564846769e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c46406e118f0fc2df2c9745310a4ad847db54442c4891d8de50798253ef79eb3
-size 809151
+oid sha256:241e9ef8782693f38efbb066bddaa7462c0e05842a834fa49d04a47ab2c067d9
+size 811269