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": 76082.8103878681,
-      "required_pages_per_second_upper_estimate": 0.006594665174300417
+      "required_token_positions_per_second": 75586.67111428632,
+      "required_pages_per_second_upper_estimate": 0.0065516610795198415
     },
     {
       "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": 9510.351298483512,
-      "required_pages_per_second_upper_estimate": 0.003793821325388349
+      "required_token_positions_per_second": 9448.33388928579,
+      "required_pages_per_second_upper_estimate": 0.003769081653616479
     },
     {
       "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": 19020.702596967025,
-      "required_pages_per_second_upper_estimate": 0.004400104792875847
+      "required_token_positions_per_second": 18896.66777857158,
+      "required_pages_per_second_upper_estimate": 0.004371411520578308
     },
     {
       "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": 28531.05389545054,
-      "required_pages_per_second_upper_estimate": 0.0017373914913618685
+      "required_token_positions_per_second": 28345.00166785737,
+      "required_pages_per_second_upper_estimate": 0.0017260618868420426
     },
     {
       "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": 28531.05389545054,
-      "required_pages_per_second_upper_estimate": 0.002196399507883847
+      "required_token_positions_per_second": 28345.00166785737,
+      "required_pages_per_second_upper_estimate": 0.0021820766923781954
     },
     {
       "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": 28531.05389545054,
-      "required_pages_per_second_upper_estimate": 0.3542602020866253
+      "required_token_positions_per_second": 28345.00166785737,
+      "required_pages_per_second_upper_estimate": 0.35195005609666824
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06262007569523576,
-    "producer_p95_seconds": 0.2119504419970326,
-    "gpu_update_seconds": 0.9690493769109824,
-    "producer_batches_per_second": 15.969319565611476,
-    "gpu_batches_per_second": 1.0319391599917005,
-    "producer_headroom": 15.47505917474816,
+    "producer_batch_seconds": 0.31023318536730926,
+    "producer_p95_seconds": 0.27335910399779095,
+    "gpu_update_seconds": 0.9754100678481259,
+    "producer_batches_per_second": 3.2233817888180534,
+    "gpu_batches_per_second": 1.025209840417295,
+    "producer_headroom": 3.144119049331431,
     "sustainable": true,
-    "buffer_seconds": 60.08106136848091,
+    "buffer_seconds": 60.4754242065838,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455298.2353808314,
+    "generation_bytes_per_second": 426459.7304848312,
     "upload_bytes_per_second": 5377766.3399886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.69640251225192,
+    "estimated_full_shards_per_hour": 91.50833068760588,
     "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": [

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 291000, "loss": 0.6449759714305401, "lr": 0.0003, "grad_norm": 0.1641612940546156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42594212353, "tokens_per_second": 150595.54317226744, "data_wait_s": 1.2327996955718845e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 291100, "loss": 0.5537864081561565, "lr": 0.0003, "grad_norm": 0.14571001136342737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42608853396, "tokens_per_second": 149374.30264801913, "data_wait_s": 1.1989999620709568e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 291200, "loss": 0.5675464998930693, "lr": 0.0003, "grad_norm": 0.16711411768102727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42623302588, "tokens_per_second": 148623.04284251956, "data_wait_s": 9.75500006461516e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 291300, "loss": 0.5636729337275028, "lr": 0.0003, "grad_norm": 0.1601071855519649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42637864279, "tokens_per_second": 148545.567102421, "data_wait_s": 1.0681000276235864e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:270cd78696fac01f736783e487574dd7f9d4ff7d2bc02554ac3ff6848093267d
-size 234711
+oid sha256:a4102de76fd2e8ce920e66b98a6b05d4b505a6f8ef3307c2db5f54bdd0144791
+size 236829