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": 76920.50371527762,
-      "required_pages_per_second_upper_estimate": 0.0053215389271812925
+      "required_token_positions_per_second": 76408.21034005903,
+      "required_pages_per_second_upper_estimate": 0.005286097282799271
     },
     {
       "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": 9615.062964409703,
-      "required_pages_per_second_upper_estimate": 0.004025626077851583
+      "required_token_positions_per_second": 9551.026292507378,
+      "required_pages_per_second_upper_estimate": 0.003998815260563851
     },
     {
       "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": 19230.125928819405,
-      "required_pages_per_second_upper_estimate": 0.004005128098449154
+      "required_token_positions_per_second": 19102.052585014757,
+      "required_pages_per_second_upper_estimate": 0.003978453798455851
     },
     {
       "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": 28845.188893229108,
-      "required_pages_per_second_upper_estimate": 0.0017292250300958785
+      "required_token_positions_per_second": 28653.078877522134,
+      "required_pages_per_second_upper_estimate": 0.0017177083279892548
     },
     {
       "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": 28845.188893229108,
-      "required_pages_per_second_upper_estimate": 0.002242586559871352
+      "required_token_positions_per_second": 28653.078877522134,
+      "required_pages_per_second_upper_estimate": 0.002227650851152791
     },
     {
       "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": 28845.188893229108,
-      "required_pages_per_second_upper_estimate": 0.3699951115715436
+      "required_token_positions_per_second": 28653.078877522134,
+      "required_pages_per_second_upper_estimate": 0.3675309305617185
     }
   ],
   "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": 0.13179256868752987,
-    "producer_p95_seconds": 0.2873201760012307,
-    "gpu_update_seconds": 0.9584960633240946,
-    "producer_batches_per_second": 7.587681232398799,
-    "gpu_batches_per_second": 1.0433011029090389,
-    "producer_headroom": 7.272762591012364,
+    "producer_batch_seconds": 0.07777828776571027,
+    "producer_p95_seconds": 0.2737700540019432,
+    "gpu_update_seconds": 0.9649224824383322,
+    "producer_batches_per_second": 12.857058553568018,
+    "gpu_batches_per_second": 1.0363526793085263,
+    "producer_headroom": 12.406064856363844,
     "sustainable": true,
-    "buffer_seconds": 60.38525198941797,
+    "buffer_seconds": 60.79011639361493,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327148.95173010696,
-    "upload_bytes_per_second": 4810880.976399813,
+    "generation_bytes_per_second": 329180.75210314227,
+    "upload_bytes_per_second": 4863670.312962996,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.19854940345198,
+    "estimated_full_shards_per_hour": 70.63452646561338,
     "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": 0.9914871209984994,
-      "bytes": 1095861,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.156548383001791,
       "bytes": 7676323,
@@ -562,6 +556,12 @@
       "bytes": 8331470,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.060381152001355,
+      "bytes": 6181629,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 949500, "loss": 0.5028177499771118, "lr": 3e-05, "grad_norm": 0.3515104382144916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138980615292, "tokens_per_second": 152106.50922828596, "data_wait_s": 1.1082000128226355e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949600, "loss": 0.5128801465034485, "lr": 3e-05, "grad_norm": 0.3471155596042175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138995110555, "tokens_per_second": 133632.28201748364, "data_wait_s": 1.266699837287888e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949700, "loss": 0.5641140341758728, "lr": 3e-05, "grad_norm": 0.3692951982825965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139009814565, "tokens_per_second": 151998.93323591017, "data_wait_s": 1.2018997949780896e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 949800, "loss": 0.5232121348381042, "lr": 3e-05, "grad_norm": 0.3377251694705753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139024570477, "tokens_per_second": 153855.84578278093, "data_wait_s": 1.2970998795935884e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47b3c795bd77f7ebb96fafa92bec0297b6e886c57b83da4dc010ec83f8790bff
-size 526443
+oid sha256:c6f7b879b4ae5ae86e2d0deb64283ecd60a4f21101a3208404cabc9fbc396c1e
+size 528561