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": 76572.4463040202,
-      "required_pages_per_second_upper_estimate": 0.0052974595078660435
+      "required_token_positions_per_second": 77248.14720117788,
+      "required_pages_per_second_upper_estimate": 0.0053442060637735
     },
     {
       "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": 9571.555788002524,
-      "required_pages_per_second_upper_estimate": 0.004007410531623053
+      "required_token_positions_per_second": 9656.018400147235,
+      "required_pages_per_second_upper_estimate": 0.004042773263548136
     },
     {
       "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": 19143.11157600505,
-      "required_pages_per_second_upper_estimate": 0.003987005303480721
+      "required_token_positions_per_second": 19312.03680029447,
+      "required_pages_per_second_upper_estimate": 0.004022187972842466
     },
     {
       "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": 28714.667364007575,
-      "required_pages_per_second_upper_estimate": 0.0017214004637138831
+      "required_token_positions_per_second": 28968.055200441708,
+      "required_pages_per_second_upper_estimate": 0.0017365906776072855
     },
     {
       "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": 28714.667364007575,
-      "required_pages_per_second_upper_estimate": 0.0022324390850779114
+      "required_token_positions_per_second": 28968.055200441708,
+      "required_pages_per_second_upper_estimate": 0.0022521388748253605
     },
     {
       "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": 28714.667364007575,
-      "required_pages_per_second_upper_estimate": 0.3683209215377891
+      "required_token_positions_per_second": 28968.055200441708,
+      "required_pages_per_second_upper_estimate": 0.37157110863690446
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0517108447811907,
-    "producer_p95_seconds": 0.17503510999995342,
-    "gpu_update_seconds": 0.9628528740909397,
-    "producer_batches_per_second": 19.33830329462612,
-    "gpu_batches_per_second": 1.0385802721356907,
-    "producer_headroom": 18.619940907273048,
+    "producer_batch_seconds": 0.072648137531246,
+    "producer_p95_seconds": 0.18127906199970312,
+    "gpu_update_seconds": 0.9544306584854347,
+    "producer_batches_per_second": 13.76497779547204,
+    "gpu_batches_per_second": 1.0477450520993095,
+    "producer_headroom": 13.137716821369766,
     "sustainable": true,
-    "buffer_seconds": 60.6597310677292,
+    "buffer_seconds": 60.12913148458239,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366747.278343058,
-    "upload_bytes_per_second": 4560657.169343154,
+    "generation_bytes_per_second": 408026.5653164203,
+    "upload_bytes_per_second": 4488655.934271871,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.69542849272543,
+    "estimated_full_shards_per_hour": 87.55300254458862,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 6176322,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4487905369996952,
+      "bytes": 925527,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 849000, "loss": 0.5520876049995422, "lr": 3e-05, "grad_norm": 0.30414955547812256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124270082823, "tokens_per_second": 152428.07822075003, "data_wait_s": 1.4350999663292896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 849100, "loss": 0.537446141242981, "lr": 3e-05, "grad_norm": 0.33703191085309125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124284744605, "tokens_per_second": 152523.14131118986, "data_wait_s": 1.2203000551380683e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 849200, "loss": 0.5582461953163147, "lr": 3e-05, "grad_norm": 0.3040351066238161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124299290318, "tokens_per_second": 151609.81058011076, "data_wait_s": 1.961599991773255e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 849300, "loss": 0.5257542133331299, "lr": 3e-05, "grad_norm": 0.29004753416830137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124313939588, "tokens_per_second": 151700.67123026052, "data_wait_s": 1.1265000466664787e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9df46787fa1737089f9894367f52d405136df3272796c4bba72ffc2da0c647d2
-size 142854
+oid sha256:c1c13d70aeac05dc8a8178bdedfca1fdeaf261bd16fd8903edbecc6001b09bee
+size 144972