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": 77112.47995290675,
-      "required_pages_per_second_upper_estimate": 0.005334820288746717
+      "required_token_positions_per_second": 76552.26440939773,
+      "required_pages_per_second_upper_estimate": 0.005296063277567614
     },
     {
       "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": 9639.059994113344,
-      "required_pages_per_second_upper_estimate": 0.004035673133073533
+      "required_token_positions_per_second": 9569.033051174716,
+      "required_pages_per_second_upper_estimate": 0.004006354314393986
     },
     {
       "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": 19278.11998822669,
-      "required_pages_per_second_upper_estimate": 0.004015123995335231
+      "required_token_positions_per_second": 19138.066102349432,
+      "required_pages_per_second_upper_estimate": 0.003985954464376344
     },
     {
       "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": 28917.179982340032,
-      "required_pages_per_second_upper_estimate": 0.0017335407859640501
+      "required_token_positions_per_second": 28707.099153524152,
+      "required_pages_per_second_upper_estimate": 0.0017209467610513902
     },
     {
       "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": 28917.179982340032,
-      "required_pages_per_second_upper_estimate": 0.0022481835503943906
+      "required_token_positions_per_second": 28707.099153524152,
+      "required_pages_per_second_upper_estimate": 0.002231850690001865
     },
     {
       "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": 28917.179982340032,
-      "required_pages_per_second_upper_estimate": 0.3709185359646494
+      "required_token_positions_per_second": 28707.099153524152,
+      "required_pages_per_second_upper_estimate": 0.36822384465981906
     }
   ],
   "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": 0.07419608516418918,
-    "producer_p95_seconds": 0.19062251899958937,
-    "gpu_update_seconds": 0.956109828720673,
-    "producer_batches_per_second": 13.477800045475323,
-    "gpu_batches_per_second": 1.0459049472779236,
-    "producer_headroom": 12.88625709301089,
+    "producer_batch_seconds": 0.370614455382821,
+    "producer_p95_seconds": 0.2203325179998501,
+    "gpu_update_seconds": 0.9631067162913208,
+    "producer_batches_per_second": 2.6982217921507243,
+    "gpu_batches_per_second": 1.0383065376708678,
+    "producer_headroom": 2.598675530063967,
     "sustainable": true,
-    "buffer_seconds": 60.2349192094024,
+    "buffer_seconds": 60.67572312635321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314035.79067980667,
-    "upload_bytes_per_second": 4812188.826838961,
+    "generation_bytes_per_second": 336562.68347330455,
+    "upload_bytes_per_second": 4791921.185196223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.38477030082369,
+    "estimated_full_shards_per_hour": 72.21851709508279,
     "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": 2.055465744000003,
-      "bytes": 6120157,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.5483520990001125,
       "bytes": 929612,
@@ -562,6 +556,12 @@
       "bytes": 6183503,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4027249899991148,
+      "bytes": 1181995,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -85,3 +85,4 @@
 {"step": 851000, "loss": 0.5287325978279114, "lr": 3e-05, "grad_norm": 0.29213773795163894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124562820078, "tokens_per_second": 151988.8861631096, "data_wait_s": 1.1604999599512666e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 851100, "loss": 0.565153956413269, "lr": 3e-05, "grad_norm": 0.3068899922873837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124577412891, "tokens_per_second": 151953.91930003065, "data_wait_s": 1.1598998753470369e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 851200, "loss": 0.5240115523338318, "lr": 3e-05, "grad_norm": 0.3030837449781713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124591975994, "tokens_per_second": 151377.53587426132, "data_wait_s": 1.4119999832473695e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 851300, "loss": 0.5254843235015869, "lr": 3e-05, "grad_norm": 0.294936959301028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124606475468, "tokens_per_second": 150537.61678752638, "data_wait_s": 8.84500150277745e-06, "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 sha256:1ef32377700322ce4e0355e021227c612ecf4623feb9ced5baa45c5985f0f371
-size 185997
+oid sha256:52eb922c739e35cba235f10976b7e6a8dcf8d2d0760808fe350251831b084bf5
+size 188115