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": 75884.03861782498,
-      "required_pages_per_second_upper_estimate": 0.006577436141055525
+      "required_token_positions_per_second": 75407.13598525718,
+      "required_pages_per_second_upper_estimate": 0.00653609942956322
     },
     {
       "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": 9485.504827228122,
-      "required_pages_per_second_upper_estimate": 0.0037839096965167234
+      "required_token_positions_per_second": 9425.891998157147,
+      "required_pages_per_second_upper_estimate": 0.0037601292477090903
     },
     {
       "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": 18971.009654456244,
-      "required_pages_per_second_upper_estimate": 0.0043886092051918955
+      "required_token_positions_per_second": 18851.783996314294,
+      "required_pages_per_second_upper_estimate": 0.0043610284474806735
     },
     {
       "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": 28456.514481684368,
-      "required_pages_per_second_upper_estimate": 0.0017328524321415886
+      "required_token_positions_per_second": 28277.67599447144,
+      "required_pages_per_second_upper_estimate": 0.0017219621065633485
     },
     {
       "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": 28456.514481684368,
-      "required_pages_per_second_upper_estimate": 0.0021906612574738235
+      "required_token_positions_per_second": 28277.67599447144,
+      "required_pages_per_second_upper_estimate": 0.0021768937756715544
     },
     {
       "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": 28456.514481684368,
-      "required_pages_per_second_upper_estimate": 0.3533346720350195
+      "required_token_positions_per_second": 28277.67599447144,
+      "required_pages_per_second_upper_estimate": 0.3511140965577491
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08476695785168431,
-    "producer_p95_seconds": 0.3149078710011963,
-    "gpu_update_seconds": 0.971587719142316,
-    "producer_batches_per_second": 11.79704952665268,
-    "gpu_batches_per_second": 1.0292431453155515,
-    "producer_headroom": 11.461868442209415,
+    "producer_batch_seconds": 0.13185721625785618,
+    "producer_p95_seconds": 0.3522655270026007,
+    "gpu_update_seconds": 0.9777323994166088,
+    "producer_batches_per_second": 7.583961108692214,
+    "gpu_batches_per_second": 1.0227747393833708,
+    "producer_headroom": 7.415084491883883,
     "sustainable": true,
-    "buffer_seconds": 60.23843858682359,
+    "buffer_seconds": 60.619408763829746,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 170,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 539084.8634763805,
-    "upload_bytes_per_second": 5711482.289431916,
+    "generation_bytes_per_second": 568218.6760800531,
+    "upload_bytes_per_second": 5737264.479859576,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.67506244867859,
+    "estimated_full_shards_per_hour": 60.96325021648977,
     "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": 0.8826371689974621,
-      "bytes": 4172226,
-      "pages": 46,
-      "shards": 3
-    },
     {
       "seconds": 2.508591282999987,
       "bytes": 5665007,
@@ -562,6 +556,12 @@
       "bytes": 10485303,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.9442589440004667,
+      "bytes": 6417253,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -250,3 +250,4 @@
 {"step": 25000, "loss": 0.7111044470220804, "lr": 0.0003, "grad_norm": 0.1274851893123072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3657994460, "tokens_per_second": 150374.07818280114, "data_wait_s": 1.0256997484248132e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 25100, "loss": 0.6755372695624828, "lr": 0.0003, "grad_norm": 0.13377661139797237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3672698039, "tokens_per_second": 138098.96691315155, "data_wait_s": 9.449999197386205e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 25200, "loss": 0.6748763881623745, "lr": 0.0003, "grad_norm": 0.14088698746413947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3687417580, "tokens_per_second": 149928.10671566988, "data_wait_s": 1.1429998266976327e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 25300, "loss": 0.6347766313701868, "lr": 0.0003, "grad_norm": 0.13177125174549342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3702096044, "tokens_per_second": 148876.27846530694, "data_wait_s": 1.0828000085894018e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be17116f3f406171f88aea523ae16f46a0211045c27ad71c600575d0b3883ffc
-size 532042
+oid sha256:09a087094077bd44481471602e4ae48674db01db6bcc6381ec298e13f5ab46da
+size 534160